1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/spinlock.h>
10 #include <linux/blkdev.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/prefetch.h>
15 #include <linux/cleancache.h>
16 #include "extent_io.h"
17 #include "extent_map.h"
19 #include "btrfs_inode.h"
21 #include "check-integrity.h"
23 #include "rcu-string.h"
27 static struct kmem_cache *extent_state_cache;
28 static struct kmem_cache *extent_buffer_cache;
29 static struct bio_set btrfs_bioset;
31 static inline bool extent_state_in_tree(const struct extent_state *state)
33 return !RB_EMPTY_NODE(&state->rb_node);
36 #ifdef CONFIG_BTRFS_DEBUG
37 static LIST_HEAD(buffers);
38 static LIST_HEAD(states);
40 static DEFINE_SPINLOCK(leak_lock);
43 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
47 spin_lock_irqsave(&leak_lock, flags);
49 spin_unlock_irqrestore(&leak_lock, flags);
53 void btrfs_leak_debug_del(struct list_head *entry)
57 spin_lock_irqsave(&leak_lock, flags);
59 spin_unlock_irqrestore(&leak_lock, flags);
63 void btrfs_leak_debug_check(void)
65 struct extent_state *state;
66 struct extent_buffer *eb;
68 while (!list_empty(&states)) {
69 state = list_entry(states.next, struct extent_state, leak_list);
70 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
71 state->start, state->end, state->state,
72 extent_state_in_tree(state),
73 refcount_read(&state->refs));
74 list_del(&state->leak_list);
75 kmem_cache_free(extent_state_cache, state);
78 while (!list_empty(&buffers)) {
79 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
80 pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n",
81 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags);
82 list_del(&eb->leak_list);
83 kmem_cache_free(extent_buffer_cache, eb);
87 #define btrfs_debug_check_extent_io_range(tree, start, end) \
88 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
89 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
90 struct extent_io_tree *tree, u64 start, u64 end)
92 struct inode *inode = tree->private_data;
95 if (!inode || !is_data_inode(inode))
98 isize = i_size_read(inode);
99 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
100 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
101 "%s: ino %llu isize %llu odd range [%llu,%llu]",
102 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
106 #define btrfs_leak_debug_add(new, head) do {} while (0)
107 #define btrfs_leak_debug_del(entry) do {} while (0)
108 #define btrfs_leak_debug_check() do {} while (0)
109 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
112 #define BUFFER_LRU_MAX 64
117 struct rb_node rb_node;
120 struct extent_page_data {
122 struct extent_io_tree *tree;
123 /* tells writepage not to lock the state bits for this range
124 * it still does the unlocking
126 unsigned int extent_locked:1;
128 /* tells the submit_bio code to use REQ_SYNC */
129 unsigned int sync_io:1;
132 static int add_extent_changeset(struct extent_state *state, unsigned bits,
133 struct extent_changeset *changeset,
140 if (set && (state->state & bits) == bits)
142 if (!set && (state->state & bits) == 0)
144 changeset->bytes_changed += state->end - state->start + 1;
145 ret = ulist_add(&changeset->range_changed, state->start, state->end,
150 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
151 unsigned long bio_flags)
153 blk_status_t ret = 0;
154 struct bio_vec *bvec = bio_last_bvec_all(bio);
155 struct page *page = bvec->bv_page;
156 struct extent_io_tree *tree = bio->bi_private;
159 start = page_offset(page) + bvec->bv_offset;
161 bio->bi_private = NULL;
164 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
165 mirror_num, bio_flags, start);
167 btrfsic_submit_bio(bio);
169 return blk_status_to_errno(ret);
172 static void flush_write_bio(struct extent_page_data *epd)
177 ret = submit_one_bio(epd->bio, 0, 0);
178 BUG_ON(ret < 0); /* -ENOMEM */
183 int __init extent_io_init(void)
185 extent_state_cache = kmem_cache_create("btrfs_extent_state",
186 sizeof(struct extent_state), 0,
187 SLAB_MEM_SPREAD, NULL);
188 if (!extent_state_cache)
191 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
192 sizeof(struct extent_buffer), 0,
193 SLAB_MEM_SPREAD, NULL);
194 if (!extent_buffer_cache)
195 goto free_state_cache;
197 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
198 offsetof(struct btrfs_io_bio, bio),
200 goto free_buffer_cache;
202 if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE))
208 bioset_exit(&btrfs_bioset);
211 kmem_cache_destroy(extent_buffer_cache);
212 extent_buffer_cache = NULL;
215 kmem_cache_destroy(extent_state_cache);
216 extent_state_cache = NULL;
220 void __cold extent_io_exit(void)
222 btrfs_leak_debug_check();
225 * Make sure all delayed rcu free are flushed before we
229 kmem_cache_destroy(extent_state_cache);
230 kmem_cache_destroy(extent_buffer_cache);
231 bioset_exit(&btrfs_bioset);
234 void extent_io_tree_init(struct extent_io_tree *tree,
237 tree->state = RB_ROOT;
239 tree->dirty_bytes = 0;
240 spin_lock_init(&tree->lock);
241 tree->private_data = private_data;
244 static struct extent_state *alloc_extent_state(gfp_t mask)
246 struct extent_state *state;
249 * The given mask might be not appropriate for the slab allocator,
250 * drop the unsupported bits
252 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
253 state = kmem_cache_alloc(extent_state_cache, mask);
257 state->failrec = NULL;
258 RB_CLEAR_NODE(&state->rb_node);
259 btrfs_leak_debug_add(&state->leak_list, &states);
260 refcount_set(&state->refs, 1);
261 init_waitqueue_head(&state->wq);
262 trace_alloc_extent_state(state, mask, _RET_IP_);
266 void free_extent_state(struct extent_state *state)
270 if (refcount_dec_and_test(&state->refs)) {
271 WARN_ON(extent_state_in_tree(state));
272 btrfs_leak_debug_del(&state->leak_list);
273 trace_free_extent_state(state, _RET_IP_);
274 kmem_cache_free(extent_state_cache, state);
278 static struct rb_node *tree_insert(struct rb_root *root,
279 struct rb_node *search_start,
281 struct rb_node *node,
282 struct rb_node ***p_in,
283 struct rb_node **parent_in)
286 struct rb_node *parent = NULL;
287 struct tree_entry *entry;
289 if (p_in && parent_in) {
295 p = search_start ? &search_start : &root->rb_node;
298 entry = rb_entry(parent, struct tree_entry, rb_node);
300 if (offset < entry->start)
302 else if (offset > entry->end)
309 rb_link_node(node, parent, p);
310 rb_insert_color(node, root);
314 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
315 struct rb_node **next_ret,
316 struct rb_node **prev_ret,
317 struct rb_node ***p_ret,
318 struct rb_node **parent_ret)
320 struct rb_root *root = &tree->state;
321 struct rb_node **n = &root->rb_node;
322 struct rb_node *prev = NULL;
323 struct rb_node *orig_prev = NULL;
324 struct tree_entry *entry;
325 struct tree_entry *prev_entry = NULL;
329 entry = rb_entry(prev, struct tree_entry, rb_node);
332 if (offset < entry->start)
334 else if (offset > entry->end)
347 while (prev && offset > prev_entry->end) {
348 prev = rb_next(prev);
349 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
356 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
357 while (prev && offset < prev_entry->start) {
358 prev = rb_prev(prev);
359 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
366 static inline struct rb_node *
367 tree_search_for_insert(struct extent_io_tree *tree,
369 struct rb_node ***p_ret,
370 struct rb_node **parent_ret)
372 struct rb_node *next= NULL;
375 ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret);
381 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
384 return tree_search_for_insert(tree, offset, NULL, NULL);
388 * utility function to look for merge candidates inside a given range.
389 * Any extents with matching state are merged together into a single
390 * extent in the tree. Extents with EXTENT_IO in their state field
391 * are not merged because the end_io handlers need to be able to do
392 * operations on them without sleeping (or doing allocations/splits).
394 * This should be called with the tree lock held.
396 static void merge_state(struct extent_io_tree *tree,
397 struct extent_state *state)
399 struct extent_state *other;
400 struct rb_node *other_node;
402 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
405 other_node = rb_prev(&state->rb_node);
407 other = rb_entry(other_node, struct extent_state, rb_node);
408 if (other->end == state->start - 1 &&
409 other->state == state->state) {
410 if (tree->private_data &&
411 is_data_inode(tree->private_data))
412 btrfs_merge_delalloc_extent(tree->private_data,
414 state->start = other->start;
415 rb_erase(&other->rb_node, &tree->state);
416 RB_CLEAR_NODE(&other->rb_node);
417 free_extent_state(other);
420 other_node = rb_next(&state->rb_node);
422 other = rb_entry(other_node, struct extent_state, rb_node);
423 if (other->start == state->end + 1 &&
424 other->state == state->state) {
425 if (tree->private_data &&
426 is_data_inode(tree->private_data))
427 btrfs_merge_delalloc_extent(tree->private_data,
429 state->end = other->end;
430 rb_erase(&other->rb_node, &tree->state);
431 RB_CLEAR_NODE(&other->rb_node);
432 free_extent_state(other);
437 static void set_state_bits(struct extent_io_tree *tree,
438 struct extent_state *state, unsigned *bits,
439 struct extent_changeset *changeset);
442 * insert an extent_state struct into the tree. 'bits' are set on the
443 * struct before it is inserted.
445 * This may return -EEXIST if the extent is already there, in which case the
446 * state struct is freed.
448 * The tree lock is not taken internally. This is a utility function and
449 * probably isn't what you want to call (see set/clear_extent_bit).
451 static int insert_state(struct extent_io_tree *tree,
452 struct extent_state *state, u64 start, u64 end,
454 struct rb_node **parent,
455 unsigned *bits, struct extent_changeset *changeset)
457 struct rb_node *node;
460 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
462 state->start = start;
465 set_state_bits(tree, state, bits, changeset);
467 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
469 struct extent_state *found;
470 found = rb_entry(node, struct extent_state, rb_node);
471 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
472 found->start, found->end, start, end);
475 merge_state(tree, state);
480 * split a given extent state struct in two, inserting the preallocated
481 * struct 'prealloc' as the newly created second half. 'split' indicates an
482 * offset inside 'orig' where it should be split.
485 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
486 * are two extent state structs in the tree:
487 * prealloc: [orig->start, split - 1]
488 * orig: [ split, orig->end ]
490 * The tree locks are not taken by this function. They need to be held
493 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
494 struct extent_state *prealloc, u64 split)
496 struct rb_node *node;
498 if (tree->private_data && is_data_inode(tree->private_data))
499 btrfs_split_delalloc_extent(tree->private_data, orig, split);
501 prealloc->start = orig->start;
502 prealloc->end = split - 1;
503 prealloc->state = orig->state;
506 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
507 &prealloc->rb_node, NULL, NULL);
509 free_extent_state(prealloc);
515 static struct extent_state *next_state(struct extent_state *state)
517 struct rb_node *next = rb_next(&state->rb_node);
519 return rb_entry(next, struct extent_state, rb_node);
525 * utility function to clear some bits in an extent state struct.
526 * it will optionally wake up anyone waiting on this state (wake == 1).
528 * If no bits are set on the state struct after clearing things, the
529 * struct is freed and removed from the tree
531 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
532 struct extent_state *state,
533 unsigned *bits, int wake,
534 struct extent_changeset *changeset)
536 struct extent_state *next;
537 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
540 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
541 u64 range = state->end - state->start + 1;
542 WARN_ON(range > tree->dirty_bytes);
543 tree->dirty_bytes -= range;
546 if (tree->private_data && is_data_inode(tree->private_data))
547 btrfs_clear_delalloc_extent(tree->private_data, state, bits);
549 ret = add_extent_changeset(state, bits_to_clear, changeset, 0);
551 state->state &= ~bits_to_clear;
554 if (state->state == 0) {
555 next = next_state(state);
556 if (extent_state_in_tree(state)) {
557 rb_erase(&state->rb_node, &tree->state);
558 RB_CLEAR_NODE(&state->rb_node);
559 free_extent_state(state);
564 merge_state(tree, state);
565 next = next_state(state);
570 static struct extent_state *
571 alloc_extent_state_atomic(struct extent_state *prealloc)
574 prealloc = alloc_extent_state(GFP_ATOMIC);
579 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
581 struct inode *inode = tree->private_data;
583 btrfs_panic(btrfs_sb(inode->i_sb), err,
584 "locking error: extent tree was modified by another thread while locked");
588 * clear some bits on a range in the tree. This may require splitting
589 * or inserting elements in the tree, so the gfp mask is used to
590 * indicate which allocations or sleeping are allowed.
592 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
593 * the given range from the tree regardless of state (ie for truncate).
595 * the range [start, end] is inclusive.
597 * This takes the tree lock, and returns 0 on success and < 0 on error.
599 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
600 unsigned bits, int wake, int delete,
601 struct extent_state **cached_state,
602 gfp_t mask, struct extent_changeset *changeset)
604 struct extent_state *state;
605 struct extent_state *cached;
606 struct extent_state *prealloc = NULL;
607 struct rb_node *node;
612 btrfs_debug_check_extent_io_range(tree, start, end);
614 if (bits & EXTENT_DELALLOC)
615 bits |= EXTENT_NORESERVE;
618 bits |= ~EXTENT_CTLBITS;
620 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
623 if (!prealloc && gfpflags_allow_blocking(mask)) {
625 * Don't care for allocation failure here because we might end
626 * up not needing the pre-allocated extent state at all, which
627 * is the case if we only have in the tree extent states that
628 * cover our input range and don't cover too any other range.
629 * If we end up needing a new extent state we allocate it later.
631 prealloc = alloc_extent_state(mask);
634 spin_lock(&tree->lock);
636 cached = *cached_state;
639 *cached_state = NULL;
643 if (cached && extent_state_in_tree(cached) &&
644 cached->start <= start && cached->end > start) {
646 refcount_dec(&cached->refs);
651 free_extent_state(cached);
654 * this search will find the extents that end after
657 node = tree_search(tree, start);
660 state = rb_entry(node, struct extent_state, rb_node);
662 if (state->start > end)
664 WARN_ON(state->end < start);
665 last_end = state->end;
667 /* the state doesn't have the wanted bits, go ahead */
668 if (!(state->state & bits)) {
669 state = next_state(state);
674 * | ---- desired range ---- |
676 * | ------------- state -------------- |
678 * We need to split the extent we found, and may flip
679 * bits on second half.
681 * If the extent we found extends past our range, we
682 * just split and search again. It'll get split again
683 * the next time though.
685 * If the extent we found is inside our range, we clear
686 * the desired bit on it.
689 if (state->start < start) {
690 prealloc = alloc_extent_state_atomic(prealloc);
692 err = split_state(tree, state, prealloc, start);
694 extent_io_tree_panic(tree, err);
699 if (state->end <= end) {
700 state = clear_state_bit(tree, state, &bits, wake,
707 * | ---- desired range ---- |
709 * We need to split the extent, and clear the bit
712 if (state->start <= end && state->end > end) {
713 prealloc = alloc_extent_state_atomic(prealloc);
715 err = split_state(tree, state, prealloc, end + 1);
717 extent_io_tree_panic(tree, err);
722 clear_state_bit(tree, prealloc, &bits, wake, changeset);
728 state = clear_state_bit(tree, state, &bits, wake, changeset);
730 if (last_end == (u64)-1)
732 start = last_end + 1;
733 if (start <= end && state && !need_resched())
739 spin_unlock(&tree->lock);
740 if (gfpflags_allow_blocking(mask))
745 spin_unlock(&tree->lock);
747 free_extent_state(prealloc);
753 static void wait_on_state(struct extent_io_tree *tree,
754 struct extent_state *state)
755 __releases(tree->lock)
756 __acquires(tree->lock)
759 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
760 spin_unlock(&tree->lock);
762 spin_lock(&tree->lock);
763 finish_wait(&state->wq, &wait);
767 * waits for one or more bits to clear on a range in the state tree.
768 * The range [start, end] is inclusive.
769 * The tree lock is taken by this function
771 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
774 struct extent_state *state;
775 struct rb_node *node;
777 btrfs_debug_check_extent_io_range(tree, start, end);
779 spin_lock(&tree->lock);
783 * this search will find all the extents that end after
786 node = tree_search(tree, start);
791 state = rb_entry(node, struct extent_state, rb_node);
793 if (state->start > end)
796 if (state->state & bits) {
797 start = state->start;
798 refcount_inc(&state->refs);
799 wait_on_state(tree, state);
800 free_extent_state(state);
803 start = state->end + 1;
808 if (!cond_resched_lock(&tree->lock)) {
809 node = rb_next(node);
814 spin_unlock(&tree->lock);
817 static void set_state_bits(struct extent_io_tree *tree,
818 struct extent_state *state,
819 unsigned *bits, struct extent_changeset *changeset)
821 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
824 if (tree->private_data && is_data_inode(tree->private_data))
825 btrfs_set_delalloc_extent(tree->private_data, state, bits);
827 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
828 u64 range = state->end - state->start + 1;
829 tree->dirty_bytes += range;
831 ret = add_extent_changeset(state, bits_to_set, changeset, 1);
833 state->state |= bits_to_set;
836 static void cache_state_if_flags(struct extent_state *state,
837 struct extent_state **cached_ptr,
840 if (cached_ptr && !(*cached_ptr)) {
841 if (!flags || (state->state & flags)) {
843 refcount_inc(&state->refs);
848 static void cache_state(struct extent_state *state,
849 struct extent_state **cached_ptr)
851 return cache_state_if_flags(state, cached_ptr,
852 EXTENT_IOBITS | EXTENT_BOUNDARY);
856 * set some bits on a range in the tree. This may require allocations or
857 * sleeping, so the gfp mask is used to indicate what is allowed.
859 * If any of the exclusive bits are set, this will fail with -EEXIST if some
860 * part of the range already has the desired bits set. The start of the
861 * existing range is returned in failed_start in this case.
863 * [start, end] is inclusive This takes the tree lock.
866 static int __must_check
867 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
868 unsigned bits, unsigned exclusive_bits,
869 u64 *failed_start, struct extent_state **cached_state,
870 gfp_t mask, struct extent_changeset *changeset)
872 struct extent_state *state;
873 struct extent_state *prealloc = NULL;
874 struct rb_node *node;
876 struct rb_node *parent;
881 btrfs_debug_check_extent_io_range(tree, start, end);
884 if (!prealloc && gfpflags_allow_blocking(mask)) {
886 * Don't care for allocation failure here because we might end
887 * up not needing the pre-allocated extent state at all, which
888 * is the case if we only have in the tree extent states that
889 * cover our input range and don't cover too any other range.
890 * If we end up needing a new extent state we allocate it later.
892 prealloc = alloc_extent_state(mask);
895 spin_lock(&tree->lock);
896 if (cached_state && *cached_state) {
897 state = *cached_state;
898 if (state->start <= start && state->end > start &&
899 extent_state_in_tree(state)) {
900 node = &state->rb_node;
905 * this search will find all the extents that end after
908 node = tree_search_for_insert(tree, start, &p, &parent);
910 prealloc = alloc_extent_state_atomic(prealloc);
912 err = insert_state(tree, prealloc, start, end,
913 &p, &parent, &bits, changeset);
915 extent_io_tree_panic(tree, err);
917 cache_state(prealloc, cached_state);
921 state = rb_entry(node, struct extent_state, rb_node);
923 last_start = state->start;
924 last_end = state->end;
927 * | ---- desired range ---- |
930 * Just lock what we found and keep going
932 if (state->start == start && state->end <= end) {
933 if (state->state & exclusive_bits) {
934 *failed_start = state->start;
939 set_state_bits(tree, state, &bits, changeset);
940 cache_state(state, cached_state);
941 merge_state(tree, state);
942 if (last_end == (u64)-1)
944 start = last_end + 1;
945 state = next_state(state);
946 if (start < end && state && state->start == start &&
953 * | ---- desired range ---- |
956 * | ------------- state -------------- |
958 * We need to split the extent we found, and may flip bits on
961 * If the extent we found extends past our
962 * range, we just split and search again. It'll get split
963 * again the next time though.
965 * If the extent we found is inside our range, we set the
968 if (state->start < start) {
969 if (state->state & exclusive_bits) {
970 *failed_start = start;
975 prealloc = alloc_extent_state_atomic(prealloc);
977 err = split_state(tree, state, prealloc, start);
979 extent_io_tree_panic(tree, err);
984 if (state->end <= end) {
985 set_state_bits(tree, state, &bits, changeset);
986 cache_state(state, cached_state);
987 merge_state(tree, state);
988 if (last_end == (u64)-1)
990 start = last_end + 1;
991 state = next_state(state);
992 if (start < end && state && state->start == start &&
999 * | ---- desired range ---- |
1000 * | state | or | state |
1002 * There's a hole, we need to insert something in it and
1003 * ignore the extent we found.
1005 if (state->start > start) {
1007 if (end < last_start)
1010 this_end = last_start - 1;
1012 prealloc = alloc_extent_state_atomic(prealloc);
1016 * Avoid to free 'prealloc' if it can be merged with
1019 err = insert_state(tree, prealloc, start, this_end,
1020 NULL, NULL, &bits, changeset);
1022 extent_io_tree_panic(tree, err);
1024 cache_state(prealloc, cached_state);
1026 start = this_end + 1;
1030 * | ---- desired range ---- |
1032 * We need to split the extent, and set the bit
1035 if (state->start <= end && state->end > end) {
1036 if (state->state & exclusive_bits) {
1037 *failed_start = start;
1042 prealloc = alloc_extent_state_atomic(prealloc);
1044 err = split_state(tree, state, prealloc, end + 1);
1046 extent_io_tree_panic(tree, err);
1048 set_state_bits(tree, prealloc, &bits, changeset);
1049 cache_state(prealloc, cached_state);
1050 merge_state(tree, prealloc);
1058 spin_unlock(&tree->lock);
1059 if (gfpflags_allow_blocking(mask))
1064 spin_unlock(&tree->lock);
1066 free_extent_state(prealloc);
1072 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1073 unsigned bits, u64 * failed_start,
1074 struct extent_state **cached_state, gfp_t mask)
1076 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1077 cached_state, mask, NULL);
1082 * convert_extent_bit - convert all bits in a given range from one bit to
1084 * @tree: the io tree to search
1085 * @start: the start offset in bytes
1086 * @end: the end offset in bytes (inclusive)
1087 * @bits: the bits to set in this range
1088 * @clear_bits: the bits to clear in this range
1089 * @cached_state: state that we're going to cache
1091 * This will go through and set bits for the given range. If any states exist
1092 * already in this range they are set with the given bit and cleared of the
1093 * clear_bits. This is only meant to be used by things that are mergeable, ie
1094 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1095 * boundary bits like LOCK.
1097 * All allocations are done with GFP_NOFS.
1099 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1100 unsigned bits, unsigned clear_bits,
1101 struct extent_state **cached_state)
1103 struct extent_state *state;
1104 struct extent_state *prealloc = NULL;
1105 struct rb_node *node;
1107 struct rb_node *parent;
1111 bool first_iteration = true;
1113 btrfs_debug_check_extent_io_range(tree, start, end);
1118 * Best effort, don't worry if extent state allocation fails
1119 * here for the first iteration. We might have a cached state
1120 * that matches exactly the target range, in which case no
1121 * extent state allocations are needed. We'll only know this
1122 * after locking the tree.
1124 prealloc = alloc_extent_state(GFP_NOFS);
1125 if (!prealloc && !first_iteration)
1129 spin_lock(&tree->lock);
1130 if (cached_state && *cached_state) {
1131 state = *cached_state;
1132 if (state->start <= start && state->end > start &&
1133 extent_state_in_tree(state)) {
1134 node = &state->rb_node;
1140 * this search will find all the extents that end after
1143 node = tree_search_for_insert(tree, start, &p, &parent);
1145 prealloc = alloc_extent_state_atomic(prealloc);
1150 err = insert_state(tree, prealloc, start, end,
1151 &p, &parent, &bits, NULL);
1153 extent_io_tree_panic(tree, err);
1154 cache_state(prealloc, cached_state);
1158 state = rb_entry(node, struct extent_state, rb_node);
1160 last_start = state->start;
1161 last_end = state->end;
1164 * | ---- desired range ---- |
1167 * Just lock what we found and keep going
1169 if (state->start == start && state->end <= end) {
1170 set_state_bits(tree, state, &bits, NULL);
1171 cache_state(state, cached_state);
1172 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1173 if (last_end == (u64)-1)
1175 start = last_end + 1;
1176 if (start < end && state && state->start == start &&
1183 * | ---- desired range ---- |
1186 * | ------------- state -------------- |
1188 * We need to split the extent we found, and may flip bits on
1191 * If the extent we found extends past our
1192 * range, we just split and search again. It'll get split
1193 * again the next time though.
1195 * If the extent we found is inside our range, we set the
1196 * desired bit on it.
1198 if (state->start < start) {
1199 prealloc = alloc_extent_state_atomic(prealloc);
1204 err = split_state(tree, state, prealloc, start);
1206 extent_io_tree_panic(tree, err);
1210 if (state->end <= end) {
1211 set_state_bits(tree, state, &bits, NULL);
1212 cache_state(state, cached_state);
1213 state = clear_state_bit(tree, state, &clear_bits, 0,
1215 if (last_end == (u64)-1)
1217 start = last_end + 1;
1218 if (start < end && state && state->start == start &&
1225 * | ---- desired range ---- |
1226 * | state | or | state |
1228 * There's a hole, we need to insert something in it and
1229 * ignore the extent we found.
1231 if (state->start > start) {
1233 if (end < last_start)
1236 this_end = last_start - 1;
1238 prealloc = alloc_extent_state_atomic(prealloc);
1245 * Avoid to free 'prealloc' if it can be merged with
1248 err = insert_state(tree, prealloc, start, this_end,
1249 NULL, NULL, &bits, NULL);
1251 extent_io_tree_panic(tree, err);
1252 cache_state(prealloc, cached_state);
1254 start = this_end + 1;
1258 * | ---- desired range ---- |
1260 * We need to split the extent, and set the bit
1263 if (state->start <= end && state->end > end) {
1264 prealloc = alloc_extent_state_atomic(prealloc);
1270 err = split_state(tree, state, prealloc, end + 1);
1272 extent_io_tree_panic(tree, err);
1274 set_state_bits(tree, prealloc, &bits, NULL);
1275 cache_state(prealloc, cached_state);
1276 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1284 spin_unlock(&tree->lock);
1286 first_iteration = false;
1290 spin_unlock(&tree->lock);
1292 free_extent_state(prealloc);
1297 /* wrappers around set/clear extent bit */
1298 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1299 unsigned bits, struct extent_changeset *changeset)
1302 * We don't support EXTENT_LOCKED yet, as current changeset will
1303 * record any bits changed, so for EXTENT_LOCKED case, it will
1304 * either fail with -EEXIST or changeset will record the whole
1307 BUG_ON(bits & EXTENT_LOCKED);
1309 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1313 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1314 unsigned bits, int wake, int delete,
1315 struct extent_state **cached)
1317 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1318 cached, GFP_NOFS, NULL);
1321 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1322 unsigned bits, struct extent_changeset *changeset)
1325 * Don't support EXTENT_LOCKED case, same reason as
1326 * set_record_extent_bits().
1328 BUG_ON(bits & EXTENT_LOCKED);
1330 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1335 * either insert or lock state struct between start and end use mask to tell
1336 * us if waiting is desired.
1338 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1339 struct extent_state **cached_state)
1345 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1346 EXTENT_LOCKED, &failed_start,
1347 cached_state, GFP_NOFS, NULL);
1348 if (err == -EEXIST) {
1349 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1350 start = failed_start;
1353 WARN_ON(start > end);
1358 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1363 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1364 &failed_start, NULL, GFP_NOFS, NULL);
1365 if (err == -EEXIST) {
1366 if (failed_start > start)
1367 clear_extent_bit(tree, start, failed_start - 1,
1368 EXTENT_LOCKED, 1, 0, NULL);
1374 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1376 unsigned long index = start >> PAGE_SHIFT;
1377 unsigned long end_index = end >> PAGE_SHIFT;
1380 while (index <= end_index) {
1381 page = find_get_page(inode->i_mapping, index);
1382 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1383 clear_page_dirty_for_io(page);
1389 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1391 unsigned long index = start >> PAGE_SHIFT;
1392 unsigned long end_index = end >> PAGE_SHIFT;
1395 while (index <= end_index) {
1396 page = find_get_page(inode->i_mapping, index);
1397 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1398 __set_page_dirty_nobuffers(page);
1399 account_page_redirty(page);
1405 /* find the first state struct with 'bits' set after 'start', and
1406 * return it. tree->lock must be held. NULL will returned if
1407 * nothing was found after 'start'
1409 static struct extent_state *
1410 find_first_extent_bit_state(struct extent_io_tree *tree,
1411 u64 start, unsigned bits)
1413 struct rb_node *node;
1414 struct extent_state *state;
1417 * this search will find all the extents that end after
1420 node = tree_search(tree, start);
1425 state = rb_entry(node, struct extent_state, rb_node);
1426 if (state->end >= start && (state->state & bits))
1429 node = rb_next(node);
1438 * find the first offset in the io tree with 'bits' set. zero is
1439 * returned if we find something, and *start_ret and *end_ret are
1440 * set to reflect the state struct that was found.
1442 * If nothing was found, 1 is returned. If found something, return 0.
1444 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1445 u64 *start_ret, u64 *end_ret, unsigned bits,
1446 struct extent_state **cached_state)
1448 struct extent_state *state;
1451 spin_lock(&tree->lock);
1452 if (cached_state && *cached_state) {
1453 state = *cached_state;
1454 if (state->end == start - 1 && extent_state_in_tree(state)) {
1455 while ((state = next_state(state)) != NULL) {
1456 if (state->state & bits)
1459 free_extent_state(*cached_state);
1460 *cached_state = NULL;
1463 free_extent_state(*cached_state);
1464 *cached_state = NULL;
1467 state = find_first_extent_bit_state(tree, start, bits);
1470 cache_state_if_flags(state, cached_state, 0);
1471 *start_ret = state->start;
1472 *end_ret = state->end;
1476 spin_unlock(&tree->lock);
1481 * find a contiguous range of bytes in the file marked as delalloc, not
1482 * more than 'max_bytes'. start and end are used to return the range,
1484 * true is returned if we find something, false if nothing was in the tree
1486 static noinline bool find_delalloc_range(struct extent_io_tree *tree,
1487 u64 *start, u64 *end, u64 max_bytes,
1488 struct extent_state **cached_state)
1490 struct rb_node *node;
1491 struct extent_state *state;
1492 u64 cur_start = *start;
1494 u64 total_bytes = 0;
1496 spin_lock(&tree->lock);
1499 * this search will find all the extents that end after
1502 node = tree_search(tree, cur_start);
1509 state = rb_entry(node, struct extent_state, rb_node);
1510 if (found && (state->start != cur_start ||
1511 (state->state & EXTENT_BOUNDARY))) {
1514 if (!(state->state & EXTENT_DELALLOC)) {
1520 *start = state->start;
1521 *cached_state = state;
1522 refcount_inc(&state->refs);
1526 cur_start = state->end + 1;
1527 node = rb_next(node);
1528 total_bytes += state->end - state->start + 1;
1529 if (total_bytes >= max_bytes)
1535 spin_unlock(&tree->lock);
1539 static int __process_pages_contig(struct address_space *mapping,
1540 struct page *locked_page,
1541 pgoff_t start_index, pgoff_t end_index,
1542 unsigned long page_ops, pgoff_t *index_ret);
1544 static noinline void __unlock_for_delalloc(struct inode *inode,
1545 struct page *locked_page,
1548 unsigned long index = start >> PAGE_SHIFT;
1549 unsigned long end_index = end >> PAGE_SHIFT;
1551 ASSERT(locked_page);
1552 if (index == locked_page->index && end_index == index)
1555 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1559 static noinline int lock_delalloc_pages(struct inode *inode,
1560 struct page *locked_page,
1564 unsigned long index = delalloc_start >> PAGE_SHIFT;
1565 unsigned long index_ret = index;
1566 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1569 ASSERT(locked_page);
1570 if (index == locked_page->index && index == end_index)
1573 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1574 end_index, PAGE_LOCK, &index_ret);
1576 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1577 (u64)index_ret << PAGE_SHIFT);
1582 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
1583 * more than @max_bytes. @Start and @end are used to return the range,
1585 * Return: true if we find something
1586 * false if nothing was in the tree
1589 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
1590 struct extent_io_tree *tree,
1591 struct page *locked_page, u64 *start,
1594 u64 max_bytes = BTRFS_MAX_EXTENT_SIZE;
1598 struct extent_state *cached_state = NULL;
1603 /* step one, find a bunch of delalloc bytes starting at start */
1604 delalloc_start = *start;
1606 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1607 max_bytes, &cached_state);
1608 if (!found || delalloc_end <= *start) {
1609 *start = delalloc_start;
1610 *end = delalloc_end;
1611 free_extent_state(cached_state);
1616 * start comes from the offset of locked_page. We have to lock
1617 * pages in order, so we can't process delalloc bytes before
1620 if (delalloc_start < *start)
1621 delalloc_start = *start;
1624 * make sure to limit the number of pages we try to lock down
1626 if (delalloc_end + 1 - delalloc_start > max_bytes)
1627 delalloc_end = delalloc_start + max_bytes - 1;
1629 /* step two, lock all the pages after the page that has start */
1630 ret = lock_delalloc_pages(inode, locked_page,
1631 delalloc_start, delalloc_end);
1632 ASSERT(!ret || ret == -EAGAIN);
1633 if (ret == -EAGAIN) {
1634 /* some of the pages are gone, lets avoid looping by
1635 * shortening the size of the delalloc range we're searching
1637 free_extent_state(cached_state);
1638 cached_state = NULL;
1640 max_bytes = PAGE_SIZE;
1649 /* step three, lock the state bits for the whole range */
1650 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1652 /* then test to make sure it is all still delalloc */
1653 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1654 EXTENT_DELALLOC, 1, cached_state);
1656 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1658 __unlock_for_delalloc(inode, locked_page,
1659 delalloc_start, delalloc_end);
1663 free_extent_state(cached_state);
1664 *start = delalloc_start;
1665 *end = delalloc_end;
1670 static int __process_pages_contig(struct address_space *mapping,
1671 struct page *locked_page,
1672 pgoff_t start_index, pgoff_t end_index,
1673 unsigned long page_ops, pgoff_t *index_ret)
1675 unsigned long nr_pages = end_index - start_index + 1;
1676 unsigned long pages_locked = 0;
1677 pgoff_t index = start_index;
1678 struct page *pages[16];
1683 if (page_ops & PAGE_LOCK) {
1684 ASSERT(page_ops == PAGE_LOCK);
1685 ASSERT(index_ret && *index_ret == start_index);
1688 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1689 mapping_set_error(mapping, -EIO);
1691 while (nr_pages > 0) {
1692 ret = find_get_pages_contig(mapping, index,
1693 min_t(unsigned long,
1694 nr_pages, ARRAY_SIZE(pages)), pages);
1697 * Only if we're going to lock these pages,
1698 * can we find nothing at @index.
1700 ASSERT(page_ops & PAGE_LOCK);
1705 for (i = 0; i < ret; i++) {
1706 if (page_ops & PAGE_SET_PRIVATE2)
1707 SetPagePrivate2(pages[i]);
1709 if (pages[i] == locked_page) {
1714 if (page_ops & PAGE_CLEAR_DIRTY)
1715 clear_page_dirty_for_io(pages[i]);
1716 if (page_ops & PAGE_SET_WRITEBACK)
1717 set_page_writeback(pages[i]);
1718 if (page_ops & PAGE_SET_ERROR)
1719 SetPageError(pages[i]);
1720 if (page_ops & PAGE_END_WRITEBACK)
1721 end_page_writeback(pages[i]);
1722 if (page_ops & PAGE_UNLOCK)
1723 unlock_page(pages[i]);
1724 if (page_ops & PAGE_LOCK) {
1725 lock_page(pages[i]);
1726 if (!PageDirty(pages[i]) ||
1727 pages[i]->mapping != mapping) {
1728 unlock_page(pages[i]);
1742 if (err && index_ret)
1743 *index_ret = start_index + pages_locked - 1;
1747 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1748 u64 delalloc_end, struct page *locked_page,
1749 unsigned clear_bits,
1750 unsigned long page_ops)
1752 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1755 __process_pages_contig(inode->i_mapping, locked_page,
1756 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1761 * count the number of bytes in the tree that have a given bit(s)
1762 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1763 * cached. The total number found is returned.
1765 u64 count_range_bits(struct extent_io_tree *tree,
1766 u64 *start, u64 search_end, u64 max_bytes,
1767 unsigned bits, int contig)
1769 struct rb_node *node;
1770 struct extent_state *state;
1771 u64 cur_start = *start;
1772 u64 total_bytes = 0;
1776 if (WARN_ON(search_end <= cur_start))
1779 spin_lock(&tree->lock);
1780 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1781 total_bytes = tree->dirty_bytes;
1785 * this search will find all the extents that end after
1788 node = tree_search(tree, cur_start);
1793 state = rb_entry(node, struct extent_state, rb_node);
1794 if (state->start > search_end)
1796 if (contig && found && state->start > last + 1)
1798 if (state->end >= cur_start && (state->state & bits) == bits) {
1799 total_bytes += min(search_end, state->end) + 1 -
1800 max(cur_start, state->start);
1801 if (total_bytes >= max_bytes)
1804 *start = max(cur_start, state->start);
1808 } else if (contig && found) {
1811 node = rb_next(node);
1816 spin_unlock(&tree->lock);
1821 * set the private field for a given byte offset in the tree. If there isn't
1822 * an extent_state there already, this does nothing.
1824 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1825 struct io_failure_record *failrec)
1827 struct rb_node *node;
1828 struct extent_state *state;
1831 spin_lock(&tree->lock);
1833 * this search will find all the extents that end after
1836 node = tree_search(tree, start);
1841 state = rb_entry(node, struct extent_state, rb_node);
1842 if (state->start != start) {
1846 state->failrec = failrec;
1848 spin_unlock(&tree->lock);
1852 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1853 struct io_failure_record **failrec)
1855 struct rb_node *node;
1856 struct extent_state *state;
1859 spin_lock(&tree->lock);
1861 * this search will find all the extents that end after
1864 node = tree_search(tree, start);
1869 state = rb_entry(node, struct extent_state, rb_node);
1870 if (state->start != start) {
1874 *failrec = state->failrec;
1876 spin_unlock(&tree->lock);
1881 * searches a range in the state tree for a given mask.
1882 * If 'filled' == 1, this returns 1 only if every extent in the tree
1883 * has the bits set. Otherwise, 1 is returned if any bit in the
1884 * range is found set.
1886 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1887 unsigned bits, int filled, struct extent_state *cached)
1889 struct extent_state *state = NULL;
1890 struct rb_node *node;
1893 spin_lock(&tree->lock);
1894 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1895 cached->end > start)
1896 node = &cached->rb_node;
1898 node = tree_search(tree, start);
1899 while (node && start <= end) {
1900 state = rb_entry(node, struct extent_state, rb_node);
1902 if (filled && state->start > start) {
1907 if (state->start > end)
1910 if (state->state & bits) {
1914 } else if (filled) {
1919 if (state->end == (u64)-1)
1922 start = state->end + 1;
1925 node = rb_next(node);
1932 spin_unlock(&tree->lock);
1937 * helper function to set a given page up to date if all the
1938 * extents in the tree for that page are up to date
1940 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1942 u64 start = page_offset(page);
1943 u64 end = start + PAGE_SIZE - 1;
1944 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1945 SetPageUptodate(page);
1948 int free_io_failure(struct extent_io_tree *failure_tree,
1949 struct extent_io_tree *io_tree,
1950 struct io_failure_record *rec)
1955 set_state_failrec(failure_tree, rec->start, NULL);
1956 ret = clear_extent_bits(failure_tree, rec->start,
1957 rec->start + rec->len - 1,
1958 EXTENT_LOCKED | EXTENT_DIRTY);
1962 ret = clear_extent_bits(io_tree, rec->start,
1963 rec->start + rec->len - 1,
1973 * this bypasses the standard btrfs submit functions deliberately, as
1974 * the standard behavior is to write all copies in a raid setup. here we only
1975 * want to write the one bad copy. so we do the mapping for ourselves and issue
1976 * submit_bio directly.
1977 * to avoid any synchronization issues, wait for the data after writing, which
1978 * actually prevents the read that triggered the error from finishing.
1979 * currently, there can be no more than two copies of every data bit. thus,
1980 * exactly one rewrite is required.
1982 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1983 u64 length, u64 logical, struct page *page,
1984 unsigned int pg_offset, int mirror_num)
1987 struct btrfs_device *dev;
1990 struct btrfs_bio *bbio = NULL;
1993 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
1994 BUG_ON(!mirror_num);
1996 bio = btrfs_io_bio_alloc(1);
1997 bio->bi_iter.bi_size = 0;
1998 map_length = length;
2001 * Avoid races with device replace and make sure our bbio has devices
2002 * associated to its stripes that don't go away while we are doing the
2003 * read repair operation.
2005 btrfs_bio_counter_inc_blocked(fs_info);
2006 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2008 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2009 * to update all raid stripes, but here we just want to correct
2010 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2011 * stripe's dev and sector.
2013 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2014 &map_length, &bbio, 0);
2016 btrfs_bio_counter_dec(fs_info);
2020 ASSERT(bbio->mirror_num == 1);
2022 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2023 &map_length, &bbio, mirror_num);
2025 btrfs_bio_counter_dec(fs_info);
2029 BUG_ON(mirror_num != bbio->mirror_num);
2032 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2033 bio->bi_iter.bi_sector = sector;
2034 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2035 btrfs_put_bbio(bbio);
2036 if (!dev || !dev->bdev ||
2037 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2038 btrfs_bio_counter_dec(fs_info);
2042 bio_set_dev(bio, dev->bdev);
2043 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2044 bio_add_page(bio, page, length, pg_offset);
2046 if (btrfsic_submit_bio_wait(bio)) {
2047 /* try to remap that extent elsewhere? */
2048 btrfs_bio_counter_dec(fs_info);
2050 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2054 btrfs_info_rl_in_rcu(fs_info,
2055 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2057 rcu_str_deref(dev->name), sector);
2058 btrfs_bio_counter_dec(fs_info);
2063 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2064 struct extent_buffer *eb, int mirror_num)
2066 u64 start = eb->start;
2067 int i, num_pages = num_extent_pages(eb);
2070 if (sb_rdonly(fs_info->sb))
2073 for (i = 0; i < num_pages; i++) {
2074 struct page *p = eb->pages[i];
2076 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2077 start - page_offset(p), mirror_num);
2087 * each time an IO finishes, we do a fast check in the IO failure tree
2088 * to see if we need to process or clean up an io_failure_record
2090 int clean_io_failure(struct btrfs_fs_info *fs_info,
2091 struct extent_io_tree *failure_tree,
2092 struct extent_io_tree *io_tree, u64 start,
2093 struct page *page, u64 ino, unsigned int pg_offset)
2096 struct io_failure_record *failrec;
2097 struct extent_state *state;
2102 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2107 ret = get_state_failrec(failure_tree, start, &failrec);
2111 BUG_ON(!failrec->this_mirror);
2113 if (failrec->in_validation) {
2114 /* there was no real error, just free the record */
2115 btrfs_debug(fs_info,
2116 "clean_io_failure: freeing dummy error at %llu",
2120 if (sb_rdonly(fs_info->sb))
2123 spin_lock(&io_tree->lock);
2124 state = find_first_extent_bit_state(io_tree,
2127 spin_unlock(&io_tree->lock);
2129 if (state && state->start <= failrec->start &&
2130 state->end >= failrec->start + failrec->len - 1) {
2131 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2133 if (num_copies > 1) {
2134 repair_io_failure(fs_info, ino, start, failrec->len,
2135 failrec->logical, page, pg_offset,
2136 failrec->failed_mirror);
2141 free_io_failure(failure_tree, io_tree, failrec);
2147 * Can be called when
2148 * - hold extent lock
2149 * - under ordered extent
2150 * - the inode is freeing
2152 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2154 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2155 struct io_failure_record *failrec;
2156 struct extent_state *state, *next;
2158 if (RB_EMPTY_ROOT(&failure_tree->state))
2161 spin_lock(&failure_tree->lock);
2162 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2164 if (state->start > end)
2167 ASSERT(state->end <= end);
2169 next = next_state(state);
2171 failrec = state->failrec;
2172 free_extent_state(state);
2177 spin_unlock(&failure_tree->lock);
2180 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2181 struct io_failure_record **failrec_ret)
2183 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2184 struct io_failure_record *failrec;
2185 struct extent_map *em;
2186 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2187 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2188 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2192 ret = get_state_failrec(failure_tree, start, &failrec);
2194 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2198 failrec->start = start;
2199 failrec->len = end - start + 1;
2200 failrec->this_mirror = 0;
2201 failrec->bio_flags = 0;
2202 failrec->in_validation = 0;
2204 read_lock(&em_tree->lock);
2205 em = lookup_extent_mapping(em_tree, start, failrec->len);
2207 read_unlock(&em_tree->lock);
2212 if (em->start > start || em->start + em->len <= start) {
2213 free_extent_map(em);
2216 read_unlock(&em_tree->lock);
2222 logical = start - em->start;
2223 logical = em->block_start + logical;
2224 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2225 logical = em->block_start;
2226 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2227 extent_set_compress_type(&failrec->bio_flags,
2231 btrfs_debug(fs_info,
2232 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2233 logical, start, failrec->len);
2235 failrec->logical = logical;
2236 free_extent_map(em);
2238 /* set the bits in the private failure tree */
2239 ret = set_extent_bits(failure_tree, start, end,
2240 EXTENT_LOCKED | EXTENT_DIRTY);
2242 ret = set_state_failrec(failure_tree, start, failrec);
2243 /* set the bits in the inode's tree */
2245 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2251 btrfs_debug(fs_info,
2252 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2253 failrec->logical, failrec->start, failrec->len,
2254 failrec->in_validation);
2256 * when data can be on disk more than twice, add to failrec here
2257 * (e.g. with a list for failed_mirror) to make
2258 * clean_io_failure() clean all those errors at once.
2262 *failrec_ret = failrec;
2267 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
2268 struct io_failure_record *failrec, int failed_mirror)
2270 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2273 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2274 if (num_copies == 1) {
2276 * we only have a single copy of the data, so don't bother with
2277 * all the retry and error correction code that follows. no
2278 * matter what the error is, it is very likely to persist.
2280 btrfs_debug(fs_info,
2281 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2282 num_copies, failrec->this_mirror, failed_mirror);
2287 * there are two premises:
2288 * a) deliver good data to the caller
2289 * b) correct the bad sectors on disk
2291 if (failed_bio_pages > 1) {
2293 * to fulfill b), we need to know the exact failing sectors, as
2294 * we don't want to rewrite any more than the failed ones. thus,
2295 * we need separate read requests for the failed bio
2297 * if the following BUG_ON triggers, our validation request got
2298 * merged. we need separate requests for our algorithm to work.
2300 BUG_ON(failrec->in_validation);
2301 failrec->in_validation = 1;
2302 failrec->this_mirror = failed_mirror;
2305 * we're ready to fulfill a) and b) alongside. get a good copy
2306 * of the failed sector and if we succeed, we have setup
2307 * everything for repair_io_failure to do the rest for us.
2309 if (failrec->in_validation) {
2310 BUG_ON(failrec->this_mirror != failed_mirror);
2311 failrec->in_validation = 0;
2312 failrec->this_mirror = 0;
2314 failrec->failed_mirror = failed_mirror;
2315 failrec->this_mirror++;
2316 if (failrec->this_mirror == failed_mirror)
2317 failrec->this_mirror++;
2320 if (failrec->this_mirror > num_copies) {
2321 btrfs_debug(fs_info,
2322 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2323 num_copies, failrec->this_mirror, failed_mirror);
2331 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2332 struct io_failure_record *failrec,
2333 struct page *page, int pg_offset, int icsum,
2334 bio_end_io_t *endio_func, void *data)
2336 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2338 struct btrfs_io_bio *btrfs_failed_bio;
2339 struct btrfs_io_bio *btrfs_bio;
2341 bio = btrfs_io_bio_alloc(1);
2342 bio->bi_end_io = endio_func;
2343 bio->bi_iter.bi_sector = failrec->logical >> 9;
2344 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2345 bio->bi_iter.bi_size = 0;
2346 bio->bi_private = data;
2348 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2349 if (btrfs_failed_bio->csum) {
2350 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2352 btrfs_bio = btrfs_io_bio(bio);
2353 btrfs_bio->csum = btrfs_bio->csum_inline;
2355 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2359 bio_add_page(bio, page, failrec->len, pg_offset);
2365 * This is a generic handler for readpage errors. If other copies exist, read
2366 * those and write back good data to the failed position. Does not investigate
2367 * in remapping the failed extent elsewhere, hoping the device will be smart
2368 * enough to do this as needed
2370 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2371 struct page *page, u64 start, u64 end,
2374 struct io_failure_record *failrec;
2375 struct inode *inode = page->mapping->host;
2376 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2377 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2380 blk_status_t status;
2382 unsigned failed_bio_pages = bio_pages_all(failed_bio);
2384 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2386 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2390 if (!btrfs_check_repairable(inode, failed_bio_pages, failrec,
2392 free_io_failure(failure_tree, tree, failrec);
2396 if (failed_bio_pages > 1)
2397 read_mode |= REQ_FAILFAST_DEV;
2399 phy_offset >>= inode->i_sb->s_blocksize_bits;
2400 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2401 start - page_offset(page),
2402 (int)phy_offset, failed_bio->bi_end_io,
2404 bio->bi_opf = REQ_OP_READ | read_mode;
2406 btrfs_debug(btrfs_sb(inode->i_sb),
2407 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2408 read_mode, failrec->this_mirror, failrec->in_validation);
2410 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2411 failrec->bio_flags, 0);
2413 free_io_failure(failure_tree, tree, failrec);
2415 ret = blk_status_to_errno(status);
2421 /* lots and lots of room for performance fixes in the end_bio funcs */
2423 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2425 int uptodate = (err == 0);
2428 btrfs_writepage_endio_finish_ordered(page, start, end, uptodate);
2431 ClearPageUptodate(page);
2433 ret = err < 0 ? err : -EIO;
2434 mapping_set_error(page->mapping, ret);
2439 * after a writepage IO is done, we need to:
2440 * clear the uptodate bits on error
2441 * clear the writeback bits in the extent tree for this IO
2442 * end_page_writeback if the page has no more pending IO
2444 * Scheduling is not allowed, so the extent state tree is expected
2445 * to have one and only one object corresponding to this IO.
2447 static void end_bio_extent_writepage(struct bio *bio)
2449 int error = blk_status_to_errno(bio->bi_status);
2450 struct bio_vec *bvec;
2455 ASSERT(!bio_flagged(bio, BIO_CLONED));
2456 bio_for_each_segment_all(bvec, bio, i) {
2457 struct page *page = bvec->bv_page;
2458 struct inode *inode = page->mapping->host;
2459 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2461 /* We always issue full-page reads, but if some block
2462 * in a page fails to read, blk_update_request() will
2463 * advance bv_offset and adjust bv_len to compensate.
2464 * Print a warning for nonzero offsets, and an error
2465 * if they don't add up to a full page. */
2466 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2467 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2469 "partial page write in btrfs with offset %u and length %u",
2470 bvec->bv_offset, bvec->bv_len);
2473 "incomplete page write in btrfs with offset %u and length %u",
2474 bvec->bv_offset, bvec->bv_len);
2477 start = page_offset(page);
2478 end = start + bvec->bv_offset + bvec->bv_len - 1;
2480 end_extent_writepage(page, error, start, end);
2481 end_page_writeback(page);
2488 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2491 struct extent_state *cached = NULL;
2492 u64 end = start + len - 1;
2494 if (uptodate && tree->track_uptodate)
2495 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2496 unlock_extent_cached_atomic(tree, start, end, &cached);
2500 * after a readpage IO is done, we need to:
2501 * clear the uptodate bits on error
2502 * set the uptodate bits if things worked
2503 * set the page up to date if all extents in the tree are uptodate
2504 * clear the lock bit in the extent tree
2505 * unlock the page if there are no other extents locked for it
2507 * Scheduling is not allowed, so the extent state tree is expected
2508 * to have one and only one object corresponding to this IO.
2510 static void end_bio_extent_readpage(struct bio *bio)
2512 struct bio_vec *bvec;
2513 int uptodate = !bio->bi_status;
2514 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2515 struct extent_io_tree *tree, *failure_tree;
2520 u64 extent_start = 0;
2526 ASSERT(!bio_flagged(bio, BIO_CLONED));
2527 bio_for_each_segment_all(bvec, bio, i) {
2528 struct page *page = bvec->bv_page;
2529 struct inode *inode = page->mapping->host;
2530 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2531 bool data_inode = btrfs_ino(BTRFS_I(inode))
2532 != BTRFS_BTREE_INODE_OBJECTID;
2534 btrfs_debug(fs_info,
2535 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2536 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2537 io_bio->mirror_num);
2538 tree = &BTRFS_I(inode)->io_tree;
2539 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2541 /* We always issue full-page reads, but if some block
2542 * in a page fails to read, blk_update_request() will
2543 * advance bv_offset and adjust bv_len to compensate.
2544 * Print a warning for nonzero offsets, and an error
2545 * if they don't add up to a full page. */
2546 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2547 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2549 "partial page read in btrfs with offset %u and length %u",
2550 bvec->bv_offset, bvec->bv_len);
2553 "incomplete page read in btrfs with offset %u and length %u",
2554 bvec->bv_offset, bvec->bv_len);
2557 start = page_offset(page);
2558 end = start + bvec->bv_offset + bvec->bv_len - 1;
2561 mirror = io_bio->mirror_num;
2562 if (likely(uptodate)) {
2563 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2569 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2570 failure_tree, tree, start,
2572 btrfs_ino(BTRFS_I(inode)), 0);
2575 if (likely(uptodate))
2581 * The generic bio_readpage_error handles errors the
2582 * following way: If possible, new read requests are
2583 * created and submitted and will end up in
2584 * end_bio_extent_readpage as well (if we're lucky,
2585 * not in the !uptodate case). In that case it returns
2586 * 0 and we just go on with the next page in our bio.
2587 * If it can't handle the error it will return -EIO and
2588 * we remain responsible for that page.
2590 ret = bio_readpage_error(bio, offset, page, start, end,
2593 uptodate = !bio->bi_status;
2598 struct extent_buffer *eb;
2600 eb = (struct extent_buffer *)page->private;
2601 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
2602 eb->read_mirror = mirror;
2603 atomic_dec(&eb->io_pages);
2604 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD,
2606 btree_readahead_hook(eb, -EIO);
2611 if (likely(uptodate)) {
2612 loff_t i_size = i_size_read(inode);
2613 pgoff_t end_index = i_size >> PAGE_SHIFT;
2616 /* Zero out the end if this page straddles i_size */
2617 off = offset_in_page(i_size);
2618 if (page->index == end_index && off)
2619 zero_user_segment(page, off, PAGE_SIZE);
2620 SetPageUptodate(page);
2622 ClearPageUptodate(page);
2628 if (unlikely(!uptodate)) {
2630 endio_readpage_release_extent(tree,
2636 endio_readpage_release_extent(tree, start,
2637 end - start + 1, 0);
2638 } else if (!extent_len) {
2639 extent_start = start;
2640 extent_len = end + 1 - start;
2641 } else if (extent_start + extent_len == start) {
2642 extent_len += end + 1 - start;
2644 endio_readpage_release_extent(tree, extent_start,
2645 extent_len, uptodate);
2646 extent_start = start;
2647 extent_len = end + 1 - start;
2652 endio_readpage_release_extent(tree, extent_start, extent_len,
2654 btrfs_io_bio_free_csum(io_bio);
2659 * Initialize the members up to but not including 'bio'. Use after allocating a
2660 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2661 * 'bio' because use of __GFP_ZERO is not supported.
2663 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2665 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2669 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2670 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2671 * for the appropriate container_of magic
2673 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2677 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset);
2678 bio_set_dev(bio, bdev);
2679 bio->bi_iter.bi_sector = first_byte >> 9;
2680 btrfs_io_bio_init(btrfs_io_bio(bio));
2684 struct bio *btrfs_bio_clone(struct bio *bio)
2686 struct btrfs_io_bio *btrfs_bio;
2689 /* Bio allocation backed by a bioset does not fail */
2690 new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset);
2691 btrfs_bio = btrfs_io_bio(new);
2692 btrfs_io_bio_init(btrfs_bio);
2693 btrfs_bio->iter = bio->bi_iter;
2697 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2701 /* Bio allocation backed by a bioset does not fail */
2702 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset);
2703 btrfs_io_bio_init(btrfs_io_bio(bio));
2707 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2710 struct btrfs_io_bio *btrfs_bio;
2712 /* this will never fail when it's backed by a bioset */
2713 bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset);
2716 btrfs_bio = btrfs_io_bio(bio);
2717 btrfs_io_bio_init(btrfs_bio);
2719 bio_trim(bio, offset >> 9, size >> 9);
2720 btrfs_bio->iter = bio->bi_iter;
2725 * @opf: bio REQ_OP_* and REQ_* flags as one value
2726 * @tree: tree so we can call our merge_bio hook
2727 * @wbc: optional writeback control for io accounting
2728 * @page: page to add to the bio
2729 * @pg_offset: offset of the new bio or to check whether we are adding
2730 * a contiguous page to the previous one
2731 * @size: portion of page that we want to write
2732 * @offset: starting offset in the page
2733 * @bdev: attach newly created bios to this bdev
2734 * @bio_ret: must be valid pointer, newly allocated bio will be stored there
2735 * @end_io_func: end_io callback for new bio
2736 * @mirror_num: desired mirror to read/write
2737 * @prev_bio_flags: flags of previous bio to see if we can merge the current one
2738 * @bio_flags: flags of the current bio to see if we can merge them
2740 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2741 struct writeback_control *wbc,
2742 struct page *page, u64 offset,
2743 size_t size, unsigned long pg_offset,
2744 struct block_device *bdev,
2745 struct bio **bio_ret,
2746 bio_end_io_t end_io_func,
2748 unsigned long prev_bio_flags,
2749 unsigned long bio_flags,
2750 bool force_bio_submit)
2754 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2755 sector_t sector = offset >> 9;
2761 bool can_merge = true;
2764 if (prev_bio_flags & EXTENT_BIO_COMPRESSED)
2765 contig = bio->bi_iter.bi_sector == sector;
2767 contig = bio_end_sector(bio) == sector;
2770 if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags))
2773 if (prev_bio_flags != bio_flags || !contig || !can_merge ||
2775 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2776 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2784 wbc_account_io(wbc, page, page_size);
2789 bio = btrfs_bio_alloc(bdev, offset);
2790 bio_add_page(bio, page, page_size, pg_offset);
2791 bio->bi_end_io = end_io_func;
2792 bio->bi_private = tree;
2793 bio->bi_write_hint = page->mapping->host->i_write_hint;
2796 wbc_init_bio(wbc, bio);
2797 wbc_account_io(wbc, page, page_size);
2805 static void attach_extent_buffer_page(struct extent_buffer *eb,
2808 if (!PagePrivate(page)) {
2809 SetPagePrivate(page);
2811 set_page_private(page, (unsigned long)eb);
2813 WARN_ON(page->private != (unsigned long)eb);
2817 void set_page_extent_mapped(struct page *page)
2819 if (!PagePrivate(page)) {
2820 SetPagePrivate(page);
2822 set_page_private(page, EXTENT_PAGE_PRIVATE);
2826 static struct extent_map *
2827 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2828 u64 start, u64 len, get_extent_t *get_extent,
2829 struct extent_map **em_cached)
2831 struct extent_map *em;
2833 if (em_cached && *em_cached) {
2835 if (extent_map_in_tree(em) && start >= em->start &&
2836 start < extent_map_end(em)) {
2837 refcount_inc(&em->refs);
2841 free_extent_map(em);
2845 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2846 if (em_cached && !IS_ERR_OR_NULL(em)) {
2848 refcount_inc(&em->refs);
2854 * basic readpage implementation. Locked extent state structs are inserted
2855 * into the tree that are removed when the IO is done (by the end_io
2857 * XXX JDM: This needs looking at to ensure proper page locking
2858 * return 0 on success, otherwise return error
2860 static int __do_readpage(struct extent_io_tree *tree,
2862 get_extent_t *get_extent,
2863 struct extent_map **em_cached,
2864 struct bio **bio, int mirror_num,
2865 unsigned long *bio_flags, unsigned int read_flags,
2868 struct inode *inode = page->mapping->host;
2869 u64 start = page_offset(page);
2870 const u64 end = start + PAGE_SIZE - 1;
2873 u64 last_byte = i_size_read(inode);
2876 struct extent_map *em;
2877 struct block_device *bdev;
2880 size_t pg_offset = 0;
2882 size_t disk_io_size;
2883 size_t blocksize = inode->i_sb->s_blocksize;
2884 unsigned long this_bio_flag = 0;
2886 set_page_extent_mapped(page);
2888 if (!PageUptodate(page)) {
2889 if (cleancache_get_page(page) == 0) {
2890 BUG_ON(blocksize != PAGE_SIZE);
2891 unlock_extent(tree, start, end);
2896 if (page->index == last_byte >> PAGE_SHIFT) {
2898 size_t zero_offset = offset_in_page(last_byte);
2901 iosize = PAGE_SIZE - zero_offset;
2902 userpage = kmap_atomic(page);
2903 memset(userpage + zero_offset, 0, iosize);
2904 flush_dcache_page(page);
2905 kunmap_atomic(userpage);
2908 while (cur <= end) {
2909 bool force_bio_submit = false;
2912 if (cur >= last_byte) {
2914 struct extent_state *cached = NULL;
2916 iosize = PAGE_SIZE - pg_offset;
2917 userpage = kmap_atomic(page);
2918 memset(userpage + pg_offset, 0, iosize);
2919 flush_dcache_page(page);
2920 kunmap_atomic(userpage);
2921 set_extent_uptodate(tree, cur, cur + iosize - 1,
2923 unlock_extent_cached(tree, cur,
2924 cur + iosize - 1, &cached);
2927 em = __get_extent_map(inode, page, pg_offset, cur,
2928 end - cur + 1, get_extent, em_cached);
2929 if (IS_ERR_OR_NULL(em)) {
2931 unlock_extent(tree, cur, end);
2934 extent_offset = cur - em->start;
2935 BUG_ON(extent_map_end(em) <= cur);
2938 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2939 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2940 extent_set_compress_type(&this_bio_flag,
2944 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2945 cur_end = min(extent_map_end(em) - 1, end);
2946 iosize = ALIGN(iosize, blocksize);
2947 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2948 disk_io_size = em->block_len;
2949 offset = em->block_start;
2951 offset = em->block_start + extent_offset;
2952 disk_io_size = iosize;
2955 block_start = em->block_start;
2956 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2957 block_start = EXTENT_MAP_HOLE;
2960 * If we have a file range that points to a compressed extent
2961 * and it's followed by a consecutive file range that points to
2962 * to the same compressed extent (possibly with a different
2963 * offset and/or length, so it either points to the whole extent
2964 * or only part of it), we must make sure we do not submit a
2965 * single bio to populate the pages for the 2 ranges because
2966 * this makes the compressed extent read zero out the pages
2967 * belonging to the 2nd range. Imagine the following scenario:
2970 * [0 - 8K] [8K - 24K]
2973 * points to extent X, points to extent X,
2974 * offset 4K, length of 8K offset 0, length 16K
2976 * [extent X, compressed length = 4K uncompressed length = 16K]
2978 * If the bio to read the compressed extent covers both ranges,
2979 * it will decompress extent X into the pages belonging to the
2980 * first range and then it will stop, zeroing out the remaining
2981 * pages that belong to the other range that points to extent X.
2982 * So here we make sure we submit 2 bios, one for the first
2983 * range and another one for the third range. Both will target
2984 * the same physical extent from disk, but we can't currently
2985 * make the compressed bio endio callback populate the pages
2986 * for both ranges because each compressed bio is tightly
2987 * coupled with a single extent map, and each range can have
2988 * an extent map with a different offset value relative to the
2989 * uncompressed data of our extent and different lengths. This
2990 * is a corner case so we prioritize correctness over
2991 * non-optimal behavior (submitting 2 bios for the same extent).
2993 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
2994 prev_em_start && *prev_em_start != (u64)-1 &&
2995 *prev_em_start != em->orig_start)
2996 force_bio_submit = true;
2999 *prev_em_start = em->orig_start;
3001 free_extent_map(em);
3004 /* we've found a hole, just zero and go on */
3005 if (block_start == EXTENT_MAP_HOLE) {
3007 struct extent_state *cached = NULL;
3009 userpage = kmap_atomic(page);
3010 memset(userpage + pg_offset, 0, iosize);
3011 flush_dcache_page(page);
3012 kunmap_atomic(userpage);
3014 set_extent_uptodate(tree, cur, cur + iosize - 1,
3016 unlock_extent_cached(tree, cur,
3017 cur + iosize - 1, &cached);
3019 pg_offset += iosize;
3022 /* the get_extent function already copied into the page */
3023 if (test_range_bit(tree, cur, cur_end,
3024 EXTENT_UPTODATE, 1, NULL)) {
3025 check_page_uptodate(tree, page);
3026 unlock_extent(tree, cur, cur + iosize - 1);
3028 pg_offset += iosize;
3031 /* we have an inline extent but it didn't get marked up
3032 * to date. Error out
3034 if (block_start == EXTENT_MAP_INLINE) {
3036 unlock_extent(tree, cur, cur + iosize - 1);
3038 pg_offset += iosize;
3042 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3043 page, offset, disk_io_size,
3044 pg_offset, bdev, bio,
3045 end_bio_extent_readpage, mirror_num,
3051 *bio_flags = this_bio_flag;
3054 unlock_extent(tree, cur, cur + iosize - 1);
3058 pg_offset += iosize;
3062 if (!PageError(page))
3063 SetPageUptodate(page);
3069 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3070 struct page *pages[], int nr_pages,
3072 struct extent_map **em_cached,
3074 unsigned long *bio_flags,
3077 struct inode *inode;
3078 struct btrfs_ordered_extent *ordered;
3081 inode = pages[0]->mapping->host;
3083 lock_extent(tree, start, end);
3084 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3088 unlock_extent(tree, start, end);
3089 btrfs_start_ordered_extent(inode, ordered, 1);
3090 btrfs_put_ordered_extent(ordered);
3093 for (index = 0; index < nr_pages; index++) {
3094 __do_readpage(tree, pages[index], btrfs_get_extent, em_cached,
3095 bio, 0, bio_flags, REQ_RAHEAD, prev_em_start);
3096 put_page(pages[index]);
3100 static void __extent_readpages(struct extent_io_tree *tree,
3101 struct page *pages[],
3103 struct extent_map **em_cached,
3104 struct bio **bio, unsigned long *bio_flags,
3111 int first_index = 0;
3113 for (index = 0; index < nr_pages; index++) {
3114 page_start = page_offset(pages[index]);
3117 end = start + PAGE_SIZE - 1;
3118 first_index = index;
3119 } else if (end + 1 == page_start) {
3122 __do_contiguous_readpages(tree, &pages[first_index],
3123 index - first_index, start,
3128 end = start + PAGE_SIZE - 1;
3129 first_index = index;
3134 __do_contiguous_readpages(tree, &pages[first_index],
3135 index - first_index, start,
3136 end, em_cached, bio,
3137 bio_flags, prev_em_start);
3140 static int __extent_read_full_page(struct extent_io_tree *tree,
3142 get_extent_t *get_extent,
3143 struct bio **bio, int mirror_num,
3144 unsigned long *bio_flags,
3145 unsigned int read_flags)
3147 struct inode *inode = page->mapping->host;
3148 struct btrfs_ordered_extent *ordered;
3149 u64 start = page_offset(page);
3150 u64 end = start + PAGE_SIZE - 1;
3154 lock_extent(tree, start, end);
3155 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3159 unlock_extent(tree, start, end);
3160 btrfs_start_ordered_extent(inode, ordered, 1);
3161 btrfs_put_ordered_extent(ordered);
3164 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3165 bio_flags, read_flags, NULL);
3169 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3170 get_extent_t *get_extent, int mirror_num)
3172 struct bio *bio = NULL;
3173 unsigned long bio_flags = 0;
3176 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3179 ret = submit_one_bio(bio, mirror_num, bio_flags);
3183 static void update_nr_written(struct writeback_control *wbc,
3184 unsigned long nr_written)
3186 wbc->nr_to_write -= nr_written;
3190 * helper for __extent_writepage, doing all of the delayed allocation setup.
3192 * This returns 1 if btrfs_run_delalloc_range function did all the work required
3193 * to write the page (copy into inline extent). In this case the IO has
3194 * been started and the page is already unlocked.
3196 * This returns 0 if all went well (page still locked)
3197 * This returns < 0 if there were errors (page still locked)
3199 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3200 struct page *page, struct writeback_control *wbc,
3201 u64 delalloc_start, unsigned long *nr_written)
3203 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
3204 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3206 u64 delalloc_to_write = 0;
3207 u64 delalloc_end = 0;
3209 int page_started = 0;
3212 while (delalloc_end < page_end) {
3213 found = find_lock_delalloc_range(inode, tree,
3218 delalloc_start = delalloc_end + 1;
3221 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
3222 delalloc_end, &page_started, nr_written, wbc);
3223 /* File system has been set read-only */
3227 * btrfs_run_delalloc_range should return < 0 for error
3228 * but just in case, we use > 0 here meaning the IO is
3229 * started, so we don't want to return > 0 unless
3230 * things are going well.
3232 ret = ret < 0 ? ret : -EIO;
3236 * delalloc_end is already one less than the total length, so
3237 * we don't subtract one from PAGE_SIZE
3239 delalloc_to_write += (delalloc_end - delalloc_start +
3240 PAGE_SIZE) >> PAGE_SHIFT;
3241 delalloc_start = delalloc_end + 1;
3243 if (wbc->nr_to_write < delalloc_to_write) {
3246 if (delalloc_to_write < thresh * 2)
3247 thresh = delalloc_to_write;
3248 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3252 /* did the fill delalloc function already unlock and start
3257 * we've unlocked the page, so we can't update
3258 * the mapping's writeback index, just update
3261 wbc->nr_to_write -= *nr_written;
3272 * helper for __extent_writepage. This calls the writepage start hooks,
3273 * and does the loop to map the page into extents and bios.
3275 * We return 1 if the IO is started and the page is unlocked,
3276 * 0 if all went well (page still locked)
3277 * < 0 if there were errors (page still locked)
3279 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3281 struct writeback_control *wbc,
3282 struct extent_page_data *epd,
3284 unsigned long nr_written,
3285 unsigned int write_flags, int *nr_ret)
3287 struct extent_io_tree *tree = epd->tree;
3288 u64 start = page_offset(page);
3289 u64 page_end = start + PAGE_SIZE - 1;
3295 struct extent_map *em;
3296 struct block_device *bdev;
3297 size_t pg_offset = 0;
3303 ret = btrfs_writepage_cow_fixup(page, start, page_end);
3305 /* Fixup worker will requeue */
3307 wbc->pages_skipped++;
3309 redirty_page_for_writepage(wbc, page);
3311 update_nr_written(wbc, nr_written);
3317 * we don't want to touch the inode after unlocking the page,
3318 * so we update the mapping writeback index now
3320 update_nr_written(wbc, nr_written + 1);
3323 if (i_size <= start) {
3324 btrfs_writepage_endio_finish_ordered(page, start, page_end, 1);
3328 blocksize = inode->i_sb->s_blocksize;
3330 while (cur <= end) {
3334 if (cur >= i_size) {
3335 btrfs_writepage_endio_finish_ordered(page, cur,
3339 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3341 if (IS_ERR_OR_NULL(em)) {
3343 ret = PTR_ERR_OR_ZERO(em);
3347 extent_offset = cur - em->start;
3348 em_end = extent_map_end(em);
3349 BUG_ON(em_end <= cur);
3351 iosize = min(em_end - cur, end - cur + 1);
3352 iosize = ALIGN(iosize, blocksize);
3353 offset = em->block_start + extent_offset;
3355 block_start = em->block_start;
3356 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3357 free_extent_map(em);
3361 * compressed and inline extents are written through other
3364 if (compressed || block_start == EXTENT_MAP_HOLE ||
3365 block_start == EXTENT_MAP_INLINE) {
3367 * end_io notification does not happen here for
3368 * compressed extents
3371 btrfs_writepage_endio_finish_ordered(page, cur,
3374 else if (compressed) {
3375 /* we don't want to end_page_writeback on
3376 * a compressed extent. this happens
3383 pg_offset += iosize;
3387 btrfs_set_range_writeback(tree, cur, cur + iosize - 1);
3388 if (!PageWriteback(page)) {
3389 btrfs_err(BTRFS_I(inode)->root->fs_info,
3390 "page %lu not writeback, cur %llu end %llu",
3391 page->index, cur, end);
3394 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3395 page, offset, iosize, pg_offset,
3397 end_bio_extent_writepage,
3401 if (PageWriteback(page))
3402 end_page_writeback(page);
3406 pg_offset += iosize;
3415 * the writepage semantics are similar to regular writepage. extent
3416 * records are inserted to lock ranges in the tree, and as dirty areas
3417 * are found, they are marked writeback. Then the lock bits are removed
3418 * and the end_io handler clears the writeback ranges
3420 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3421 struct extent_page_data *epd)
3423 struct inode *inode = page->mapping->host;
3424 u64 start = page_offset(page);
3425 u64 page_end = start + PAGE_SIZE - 1;
3428 size_t pg_offset = 0;
3429 loff_t i_size = i_size_read(inode);
3430 unsigned long end_index = i_size >> PAGE_SHIFT;
3431 unsigned int write_flags = 0;
3432 unsigned long nr_written = 0;
3434 write_flags = wbc_to_write_flags(wbc);
3436 trace___extent_writepage(page, inode, wbc);
3438 WARN_ON(!PageLocked(page));
3440 ClearPageError(page);
3442 pg_offset = offset_in_page(i_size);
3443 if (page->index > end_index ||
3444 (page->index == end_index && !pg_offset)) {
3445 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3450 if (page->index == end_index) {
3453 userpage = kmap_atomic(page);
3454 memset(userpage + pg_offset, 0,
3455 PAGE_SIZE - pg_offset);
3456 kunmap_atomic(userpage);
3457 flush_dcache_page(page);
3462 set_page_extent_mapped(page);
3464 if (!epd->extent_locked) {
3465 ret = writepage_delalloc(inode, page, wbc, start, &nr_written);
3472 ret = __extent_writepage_io(inode, page, wbc, epd,
3473 i_size, nr_written, write_flags, &nr);
3479 /* make sure the mapping tag for page dirty gets cleared */
3480 set_page_writeback(page);
3481 end_page_writeback(page);
3483 if (PageError(page)) {
3484 ret = ret < 0 ? ret : -EIO;
3485 end_extent_writepage(page, ret, start, page_end);
3494 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3496 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3497 TASK_UNINTERRUPTIBLE);
3500 static noinline_for_stack int
3501 lock_extent_buffer_for_io(struct extent_buffer *eb,
3502 struct btrfs_fs_info *fs_info,
3503 struct extent_page_data *epd)
3509 if (!btrfs_try_tree_write_lock(eb)) {
3511 flush_write_bio(epd);
3512 btrfs_tree_lock(eb);
3515 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3516 btrfs_tree_unlock(eb);
3520 flush_write_bio(epd);
3524 wait_on_extent_buffer_writeback(eb);
3525 btrfs_tree_lock(eb);
3526 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3528 btrfs_tree_unlock(eb);
3533 * We need to do this to prevent races in people who check if the eb is
3534 * under IO since we can end up having no IO bits set for a short period
3537 spin_lock(&eb->refs_lock);
3538 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3539 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3540 spin_unlock(&eb->refs_lock);
3541 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3542 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3544 fs_info->dirty_metadata_batch);
3547 spin_unlock(&eb->refs_lock);
3550 btrfs_tree_unlock(eb);
3555 num_pages = num_extent_pages(eb);
3556 for (i = 0; i < num_pages; i++) {
3557 struct page *p = eb->pages[i];
3559 if (!trylock_page(p)) {
3561 flush_write_bio(epd);
3571 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3573 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3574 smp_mb__after_atomic();
3575 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3578 static void set_btree_ioerr(struct page *page)
3580 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3583 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3587 * If writeback for a btree extent that doesn't belong to a log tree
3588 * failed, increment the counter transaction->eb_write_errors.
3589 * We do this because while the transaction is running and before it's
3590 * committing (when we call filemap_fdata[write|wait]_range against
3591 * the btree inode), we might have
3592 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3593 * returns an error or an error happens during writeback, when we're
3594 * committing the transaction we wouldn't know about it, since the pages
3595 * can be no longer dirty nor marked anymore for writeback (if a
3596 * subsequent modification to the extent buffer didn't happen before the
3597 * transaction commit), which makes filemap_fdata[write|wait]_range not
3598 * able to find the pages tagged with SetPageError at transaction
3599 * commit time. So if this happens we must abort the transaction,
3600 * otherwise we commit a super block with btree roots that point to
3601 * btree nodes/leafs whose content on disk is invalid - either garbage
3602 * or the content of some node/leaf from a past generation that got
3603 * cowed or deleted and is no longer valid.
3605 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3606 * not be enough - we need to distinguish between log tree extents vs
3607 * non-log tree extents, and the next filemap_fdatawait_range() call
3608 * will catch and clear such errors in the mapping - and that call might
3609 * be from a log sync and not from a transaction commit. Also, checking
3610 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3611 * not done and would not be reliable - the eb might have been released
3612 * from memory and reading it back again means that flag would not be
3613 * set (since it's a runtime flag, not persisted on disk).
3615 * Using the flags below in the btree inode also makes us achieve the
3616 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3617 * writeback for all dirty pages and before filemap_fdatawait_range()
3618 * is called, the writeback for all dirty pages had already finished
3619 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3620 * filemap_fdatawait_range() would return success, as it could not know
3621 * that writeback errors happened (the pages were no longer tagged for
3624 switch (eb->log_index) {
3626 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3629 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3632 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3635 BUG(); /* unexpected, logic error */
3639 static void end_bio_extent_buffer_writepage(struct bio *bio)
3641 struct bio_vec *bvec;
3642 struct extent_buffer *eb;
3645 ASSERT(!bio_flagged(bio, BIO_CLONED));
3646 bio_for_each_segment_all(bvec, bio, i) {
3647 struct page *page = bvec->bv_page;
3649 eb = (struct extent_buffer *)page->private;
3651 done = atomic_dec_and_test(&eb->io_pages);
3653 if (bio->bi_status ||
3654 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3655 ClearPageUptodate(page);
3656 set_btree_ioerr(page);
3659 end_page_writeback(page);
3664 end_extent_buffer_writeback(eb);
3670 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3671 struct btrfs_fs_info *fs_info,
3672 struct writeback_control *wbc,
3673 struct extent_page_data *epd)
3675 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3676 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3677 u64 offset = eb->start;
3680 unsigned long start, end;
3681 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3684 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3685 num_pages = num_extent_pages(eb);
3686 atomic_set(&eb->io_pages, num_pages);
3688 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3689 nritems = btrfs_header_nritems(eb);
3690 if (btrfs_header_level(eb) > 0) {
3691 end = btrfs_node_key_ptr_offset(nritems);
3693 memzero_extent_buffer(eb, end, eb->len - end);
3697 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3699 start = btrfs_item_nr_offset(nritems);
3700 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3701 memzero_extent_buffer(eb, start, end - start);
3704 for (i = 0; i < num_pages; i++) {
3705 struct page *p = eb->pages[i];
3707 clear_page_dirty_for_io(p);
3708 set_page_writeback(p);
3709 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3710 p, offset, PAGE_SIZE, 0, bdev,
3712 end_bio_extent_buffer_writepage,
3716 if (PageWriteback(p))
3717 end_page_writeback(p);
3718 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3719 end_extent_buffer_writeback(eb);
3723 offset += PAGE_SIZE;
3724 update_nr_written(wbc, 1);
3728 if (unlikely(ret)) {
3729 for (; i < num_pages; i++) {
3730 struct page *p = eb->pages[i];
3731 clear_page_dirty_for_io(p);
3739 int btree_write_cache_pages(struct address_space *mapping,
3740 struct writeback_control *wbc)
3742 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3743 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3744 struct extent_buffer *eb, *prev_eb = NULL;
3745 struct extent_page_data epd = {
3749 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3753 int nr_to_write_done = 0;
3754 struct pagevec pvec;
3757 pgoff_t end; /* Inclusive */
3761 pagevec_init(&pvec);
3762 if (wbc->range_cyclic) {
3763 index = mapping->writeback_index; /* Start from prev offset */
3766 index = wbc->range_start >> PAGE_SHIFT;
3767 end = wbc->range_end >> PAGE_SHIFT;
3770 if (wbc->sync_mode == WB_SYNC_ALL)
3771 tag = PAGECACHE_TAG_TOWRITE;
3773 tag = PAGECACHE_TAG_DIRTY;
3775 if (wbc->sync_mode == WB_SYNC_ALL)
3776 tag_pages_for_writeback(mapping, index, end);
3777 while (!done && !nr_to_write_done && (index <= end) &&
3778 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3783 for (i = 0; i < nr_pages; i++) {
3784 struct page *page = pvec.pages[i];
3786 if (!PagePrivate(page))
3789 spin_lock(&mapping->private_lock);
3790 if (!PagePrivate(page)) {
3791 spin_unlock(&mapping->private_lock);
3795 eb = (struct extent_buffer *)page->private;
3798 * Shouldn't happen and normally this would be a BUG_ON
3799 * but no sense in crashing the users box for something
3800 * we can survive anyway.
3803 spin_unlock(&mapping->private_lock);
3807 if (eb == prev_eb) {
3808 spin_unlock(&mapping->private_lock);
3812 ret = atomic_inc_not_zero(&eb->refs);
3813 spin_unlock(&mapping->private_lock);
3818 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3820 free_extent_buffer(eb);
3824 ret = write_one_eb(eb, fs_info, wbc, &epd);
3827 free_extent_buffer(eb);
3830 free_extent_buffer(eb);
3833 * the filesystem may choose to bump up nr_to_write.
3834 * We have to make sure to honor the new nr_to_write
3837 nr_to_write_done = wbc->nr_to_write <= 0;
3839 pagevec_release(&pvec);
3842 if (!scanned && !done) {
3844 * We hit the last page and there is more work to be done: wrap
3845 * back to the start of the file
3851 flush_write_bio(&epd);
3856 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3857 * @mapping: address space structure to write
3858 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3859 * @data: data passed to __extent_writepage function
3861 * If a page is already under I/O, write_cache_pages() skips it, even
3862 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3863 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3864 * and msync() need to guarantee that all the data which was dirty at the time
3865 * the call was made get new I/O started against them. If wbc->sync_mode is
3866 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3867 * existing IO to complete.
3869 static int extent_write_cache_pages(struct address_space *mapping,
3870 struct writeback_control *wbc,
3871 struct extent_page_data *epd)
3873 struct inode *inode = mapping->host;
3876 int nr_to_write_done = 0;
3877 struct pagevec pvec;
3880 pgoff_t end; /* Inclusive */
3882 int range_whole = 0;
3887 * We have to hold onto the inode so that ordered extents can do their
3888 * work when the IO finishes. The alternative to this is failing to add
3889 * an ordered extent if the igrab() fails there and that is a huge pain
3890 * to deal with, so instead just hold onto the inode throughout the
3891 * writepages operation. If it fails here we are freeing up the inode
3892 * anyway and we'd rather not waste our time writing out stuff that is
3893 * going to be truncated anyway.
3898 pagevec_init(&pvec);
3899 if (wbc->range_cyclic) {
3900 index = mapping->writeback_index; /* Start from prev offset */
3903 index = wbc->range_start >> PAGE_SHIFT;
3904 end = wbc->range_end >> PAGE_SHIFT;
3905 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3911 * We do the tagged writepage as long as the snapshot flush bit is set
3912 * and we are the first one who do the filemap_flush() on this inode.
3914 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
3915 * not race in and drop the bit.
3917 if (range_whole && wbc->nr_to_write == LONG_MAX &&
3918 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
3919 &BTRFS_I(inode)->runtime_flags))
3920 wbc->tagged_writepages = 1;
3922 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3923 tag = PAGECACHE_TAG_TOWRITE;
3925 tag = PAGECACHE_TAG_DIRTY;
3927 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3928 tag_pages_for_writeback(mapping, index, end);
3930 while (!done && !nr_to_write_done && (index <= end) &&
3931 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3932 &index, end, tag))) {
3936 for (i = 0; i < nr_pages; i++) {
3937 struct page *page = pvec.pages[i];
3939 done_index = page->index;
3941 * At this point we hold neither the i_pages lock nor
3942 * the page lock: the page may be truncated or
3943 * invalidated (changing page->mapping to NULL),
3944 * or even swizzled back from swapper_space to
3945 * tmpfs file mapping
3947 if (!trylock_page(page)) {
3948 flush_write_bio(epd);
3952 if (unlikely(page->mapping != mapping)) {
3957 if (wbc->sync_mode != WB_SYNC_NONE) {
3958 if (PageWriteback(page))
3959 flush_write_bio(epd);
3960 wait_on_page_writeback(page);
3963 if (PageWriteback(page) ||
3964 !clear_page_dirty_for_io(page)) {
3969 ret = __extent_writepage(page, wbc, epd);
3971 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3977 * done_index is set past this page,
3978 * so media errors will not choke
3979 * background writeout for the entire
3980 * file. This has consequences for
3981 * range_cyclic semantics (ie. it may
3982 * not be suitable for data integrity
3985 done_index = page->index + 1;
3991 * the filesystem may choose to bump up nr_to_write.
3992 * We have to make sure to honor the new nr_to_write
3995 nr_to_write_done = wbc->nr_to_write <= 0;
3997 pagevec_release(&pvec);
4000 if (!scanned && !done) {
4002 * We hit the last page and there is more work to be done: wrap
4003 * back to the start of the file
4010 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4011 mapping->writeback_index = done_index;
4013 btrfs_add_delayed_iput(inode);
4017 int extent_write_full_page(struct page *page, struct writeback_control *wbc)
4020 struct extent_page_data epd = {
4022 .tree = &BTRFS_I(page->mapping->host)->io_tree,
4024 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4027 ret = __extent_writepage(page, wbc, &epd);
4029 flush_write_bio(&epd);
4033 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
4037 struct address_space *mapping = inode->i_mapping;
4038 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
4040 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4043 struct extent_page_data epd = {
4047 .sync_io = mode == WB_SYNC_ALL,
4049 struct writeback_control wbc_writepages = {
4051 .nr_to_write = nr_pages * 2,
4052 .range_start = start,
4053 .range_end = end + 1,
4056 while (start <= end) {
4057 page = find_get_page(mapping, start >> PAGE_SHIFT);
4058 if (clear_page_dirty_for_io(page))
4059 ret = __extent_writepage(page, &wbc_writepages, &epd);
4061 btrfs_writepage_endio_finish_ordered(page, start,
4062 start + PAGE_SIZE - 1, 1);
4069 flush_write_bio(&epd);
4073 int extent_writepages(struct address_space *mapping,
4074 struct writeback_control *wbc)
4077 struct extent_page_data epd = {
4079 .tree = &BTRFS_I(mapping->host)->io_tree,
4081 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4084 ret = extent_write_cache_pages(mapping, wbc, &epd);
4085 flush_write_bio(&epd);
4089 int extent_readpages(struct address_space *mapping, struct list_head *pages,
4092 struct bio *bio = NULL;
4093 unsigned long bio_flags = 0;
4094 struct page *pagepool[16];
4095 struct extent_map *em_cached = NULL;
4096 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
4098 u64 prev_em_start = (u64)-1;
4100 while (!list_empty(pages)) {
4101 for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) {
4102 struct page *page = lru_to_page(pages);
4104 prefetchw(&page->flags);
4105 list_del(&page->lru);
4106 if (add_to_page_cache_lru(page, mapping, page->index,
4107 readahead_gfp_mask(mapping))) {
4112 pagepool[nr++] = page;
4115 __extent_readpages(tree, pagepool, nr, &em_cached, &bio,
4116 &bio_flags, &prev_em_start);
4120 free_extent_map(em_cached);
4123 return submit_one_bio(bio, 0, bio_flags);
4128 * basic invalidatepage code, this waits on any locked or writeback
4129 * ranges corresponding to the page, and then deletes any extent state
4130 * records from the tree
4132 int extent_invalidatepage(struct extent_io_tree *tree,
4133 struct page *page, unsigned long offset)
4135 struct extent_state *cached_state = NULL;
4136 u64 start = page_offset(page);
4137 u64 end = start + PAGE_SIZE - 1;
4138 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4140 start += ALIGN(offset, blocksize);
4144 lock_extent_bits(tree, start, end, &cached_state);
4145 wait_on_page_writeback(page);
4146 clear_extent_bit(tree, start, end,
4147 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4148 EXTENT_DO_ACCOUNTING,
4149 1, 1, &cached_state);
4154 * a helper for releasepage, this tests for areas of the page that
4155 * are locked or under IO and drops the related state bits if it is safe
4158 static int try_release_extent_state(struct extent_io_tree *tree,
4159 struct page *page, gfp_t mask)
4161 u64 start = page_offset(page);
4162 u64 end = start + PAGE_SIZE - 1;
4165 if (test_range_bit(tree, start, end,
4166 EXTENT_IOBITS, 0, NULL))
4170 * at this point we can safely clear everything except the
4171 * locked bit and the nodatasum bit
4173 ret = __clear_extent_bit(tree, start, end,
4174 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4175 0, 0, NULL, mask, NULL);
4177 /* if clear_extent_bit failed for enomem reasons,
4178 * we can't allow the release to continue.
4189 * a helper for releasepage. As long as there are no locked extents
4190 * in the range corresponding to the page, both state records and extent
4191 * map records are removed
4193 int try_release_extent_mapping(struct page *page, gfp_t mask)
4195 struct extent_map *em;
4196 u64 start = page_offset(page);
4197 u64 end = start + PAGE_SIZE - 1;
4198 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4199 struct extent_io_tree *tree = &btrfs_inode->io_tree;
4200 struct extent_map_tree *map = &btrfs_inode->extent_tree;
4202 if (gfpflags_allow_blocking(mask) &&
4203 page->mapping->host->i_size > SZ_16M) {
4205 while (start <= end) {
4206 len = end - start + 1;
4207 write_lock(&map->lock);
4208 em = lookup_extent_mapping(map, start, len);
4210 write_unlock(&map->lock);
4213 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4214 em->start != start) {
4215 write_unlock(&map->lock);
4216 free_extent_map(em);
4219 if (!test_range_bit(tree, em->start,
4220 extent_map_end(em) - 1,
4221 EXTENT_LOCKED | EXTENT_WRITEBACK,
4223 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4224 &btrfs_inode->runtime_flags);
4225 remove_extent_mapping(map, em);
4226 /* once for the rb tree */
4227 free_extent_map(em);
4229 start = extent_map_end(em);
4230 write_unlock(&map->lock);
4233 free_extent_map(em);
4236 return try_release_extent_state(tree, page, mask);
4240 * helper function for fiemap, which doesn't want to see any holes.
4241 * This maps until we find something past 'last'
4243 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4244 u64 offset, u64 last)
4246 u64 sectorsize = btrfs_inode_sectorsize(inode);
4247 struct extent_map *em;
4254 len = last - offset;
4257 len = ALIGN(len, sectorsize);
4258 em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len);
4259 if (IS_ERR_OR_NULL(em))
4262 /* if this isn't a hole return it */
4263 if (em->block_start != EXTENT_MAP_HOLE)
4266 /* this is a hole, advance to the next extent */
4267 offset = extent_map_end(em);
4268 free_extent_map(em);
4276 * To cache previous fiemap extent
4278 * Will be used for merging fiemap extent
4280 struct fiemap_cache {
4289 * Helper to submit fiemap extent.
4291 * Will try to merge current fiemap extent specified by @offset, @phys,
4292 * @len and @flags with cached one.
4293 * And only when we fails to merge, cached one will be submitted as
4296 * Return value is the same as fiemap_fill_next_extent().
4298 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4299 struct fiemap_cache *cache,
4300 u64 offset, u64 phys, u64 len, u32 flags)
4308 * Sanity check, extent_fiemap() should have ensured that new
4309 * fiemap extent won't overlap with cached one.
4312 * NOTE: Physical address can overlap, due to compression
4314 if (cache->offset + cache->len > offset) {
4320 * Only merges fiemap extents if
4321 * 1) Their logical addresses are continuous
4323 * 2) Their physical addresses are continuous
4324 * So truly compressed (physical size smaller than logical size)
4325 * extents won't get merged with each other
4327 * 3) Share same flags except FIEMAP_EXTENT_LAST
4328 * So regular extent won't get merged with prealloc extent
4330 if (cache->offset + cache->len == offset &&
4331 cache->phys + cache->len == phys &&
4332 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4333 (flags & ~FIEMAP_EXTENT_LAST)) {
4335 cache->flags |= flags;
4336 goto try_submit_last;
4339 /* Not mergeable, need to submit cached one */
4340 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4341 cache->len, cache->flags);
4342 cache->cached = false;
4346 cache->cached = true;
4347 cache->offset = offset;
4350 cache->flags = flags;
4352 if (cache->flags & FIEMAP_EXTENT_LAST) {
4353 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4354 cache->phys, cache->len, cache->flags);
4355 cache->cached = false;
4361 * Emit last fiemap cache
4363 * The last fiemap cache may still be cached in the following case:
4365 * |<- Fiemap range ->|
4366 * |<------------ First extent ----------->|
4368 * In this case, the first extent range will be cached but not emitted.
4369 * So we must emit it before ending extent_fiemap().
4371 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4372 struct fiemap_extent_info *fieinfo,
4373 struct fiemap_cache *cache)
4380 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4381 cache->len, cache->flags);
4382 cache->cached = false;
4388 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4389 __u64 start, __u64 len)
4393 u64 max = start + len;
4397 u64 last_for_get_extent = 0;
4399 u64 isize = i_size_read(inode);
4400 struct btrfs_key found_key;
4401 struct extent_map *em = NULL;
4402 struct extent_state *cached_state = NULL;
4403 struct btrfs_path *path;
4404 struct btrfs_root *root = BTRFS_I(inode)->root;
4405 struct fiemap_cache cache = { 0 };
4414 path = btrfs_alloc_path();
4417 path->leave_spinning = 1;
4419 start = round_down(start, btrfs_inode_sectorsize(inode));
4420 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4423 * lookup the last file extent. We're not using i_size here
4424 * because there might be preallocation past i_size
4426 ret = btrfs_lookup_file_extent(NULL, root, path,
4427 btrfs_ino(BTRFS_I(inode)), -1, 0);
4429 btrfs_free_path(path);
4438 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4439 found_type = found_key.type;
4441 /* No extents, but there might be delalloc bits */
4442 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4443 found_type != BTRFS_EXTENT_DATA_KEY) {
4444 /* have to trust i_size as the end */
4446 last_for_get_extent = isize;
4449 * remember the start of the last extent. There are a
4450 * bunch of different factors that go into the length of the
4451 * extent, so its much less complex to remember where it started
4453 last = found_key.offset;
4454 last_for_get_extent = last + 1;
4456 btrfs_release_path(path);
4459 * we might have some extents allocated but more delalloc past those
4460 * extents. so, we trust isize unless the start of the last extent is
4465 last_for_get_extent = isize;
4468 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4471 em = get_extent_skip_holes(inode, start, last_for_get_extent);
4480 u64 offset_in_extent = 0;
4482 /* break if the extent we found is outside the range */
4483 if (em->start >= max || extent_map_end(em) < off)
4487 * get_extent may return an extent that starts before our
4488 * requested range. We have to make sure the ranges
4489 * we return to fiemap always move forward and don't
4490 * overlap, so adjust the offsets here
4492 em_start = max(em->start, off);
4495 * record the offset from the start of the extent
4496 * for adjusting the disk offset below. Only do this if the
4497 * extent isn't compressed since our in ram offset may be past
4498 * what we have actually allocated on disk.
4500 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4501 offset_in_extent = em_start - em->start;
4502 em_end = extent_map_end(em);
4503 em_len = em_end - em_start;
4505 if (em->block_start < EXTENT_MAP_LAST_BYTE)
4506 disko = em->block_start + offset_in_extent;
4511 * bump off for our next call to get_extent
4513 off = extent_map_end(em);
4517 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4519 flags |= FIEMAP_EXTENT_LAST;
4520 } else if (em->block_start == EXTENT_MAP_INLINE) {
4521 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4522 FIEMAP_EXTENT_NOT_ALIGNED);
4523 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4524 flags |= (FIEMAP_EXTENT_DELALLOC |
4525 FIEMAP_EXTENT_UNKNOWN);
4526 } else if (fieinfo->fi_extents_max) {
4527 u64 bytenr = em->block_start -
4528 (em->start - em->orig_start);
4531 * As btrfs supports shared space, this information
4532 * can be exported to userspace tools via
4533 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4534 * then we're just getting a count and we can skip the
4537 ret = btrfs_check_shared(root,
4538 btrfs_ino(BTRFS_I(inode)),
4543 flags |= FIEMAP_EXTENT_SHARED;
4546 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4547 flags |= FIEMAP_EXTENT_ENCODED;
4548 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4549 flags |= FIEMAP_EXTENT_UNWRITTEN;
4551 free_extent_map(em);
4553 if ((em_start >= last) || em_len == (u64)-1 ||
4554 (last == (u64)-1 && isize <= em_end)) {
4555 flags |= FIEMAP_EXTENT_LAST;
4559 /* now scan forward to see if this is really the last extent. */
4560 em = get_extent_skip_holes(inode, off, last_for_get_extent);
4566 flags |= FIEMAP_EXTENT_LAST;
4569 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4579 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4580 free_extent_map(em);
4582 btrfs_free_path(path);
4583 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4588 static void __free_extent_buffer(struct extent_buffer *eb)
4590 btrfs_leak_debug_del(&eb->leak_list);
4591 kmem_cache_free(extent_buffer_cache, eb);
4594 int extent_buffer_under_io(struct extent_buffer *eb)
4596 return (atomic_read(&eb->io_pages) ||
4597 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4598 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4602 * Release all pages attached to the extent buffer.
4604 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4608 int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4610 BUG_ON(extent_buffer_under_io(eb));
4612 num_pages = num_extent_pages(eb);
4613 for (i = 0; i < num_pages; i++) {
4614 struct page *page = eb->pages[i];
4619 spin_lock(&page->mapping->private_lock);
4621 * We do this since we'll remove the pages after we've
4622 * removed the eb from the radix tree, so we could race
4623 * and have this page now attached to the new eb. So
4624 * only clear page_private if it's still connected to
4627 if (PagePrivate(page) &&
4628 page->private == (unsigned long)eb) {
4629 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4630 BUG_ON(PageDirty(page));
4631 BUG_ON(PageWriteback(page));
4633 * We need to make sure we haven't be attached
4636 ClearPagePrivate(page);
4637 set_page_private(page, 0);
4638 /* One for the page private */
4643 spin_unlock(&page->mapping->private_lock);
4645 /* One for when we allocated the page */
4651 * Helper for releasing the extent buffer.
4653 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4655 btrfs_release_extent_buffer_pages(eb);
4656 __free_extent_buffer(eb);
4659 static struct extent_buffer *
4660 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4663 struct extent_buffer *eb = NULL;
4665 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4668 eb->fs_info = fs_info;
4670 rwlock_init(&eb->lock);
4671 atomic_set(&eb->write_locks, 0);
4672 atomic_set(&eb->read_locks, 0);
4673 atomic_set(&eb->blocking_readers, 0);
4674 atomic_set(&eb->blocking_writers, 0);
4675 atomic_set(&eb->spinning_readers, 0);
4676 atomic_set(&eb->spinning_writers, 0);
4677 eb->lock_nested = 0;
4678 init_waitqueue_head(&eb->write_lock_wq);
4679 init_waitqueue_head(&eb->read_lock_wq);
4681 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4683 spin_lock_init(&eb->refs_lock);
4684 atomic_set(&eb->refs, 1);
4685 atomic_set(&eb->io_pages, 0);
4688 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4690 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4691 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4692 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4697 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4701 struct extent_buffer *new;
4702 int num_pages = num_extent_pages(src);
4704 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4708 for (i = 0; i < num_pages; i++) {
4709 p = alloc_page(GFP_NOFS);
4711 btrfs_release_extent_buffer(new);
4714 attach_extent_buffer_page(new, p);
4715 WARN_ON(PageDirty(p));
4718 copy_page(page_address(p), page_address(src->pages[i]));
4721 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4722 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4727 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4728 u64 start, unsigned long len)
4730 struct extent_buffer *eb;
4734 eb = __alloc_extent_buffer(fs_info, start, len);
4738 num_pages = num_extent_pages(eb);
4739 for (i = 0; i < num_pages; i++) {
4740 eb->pages[i] = alloc_page(GFP_NOFS);
4744 set_extent_buffer_uptodate(eb);
4745 btrfs_set_header_nritems(eb, 0);
4746 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4751 __free_page(eb->pages[i - 1]);
4752 __free_extent_buffer(eb);
4756 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4759 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4762 static void check_buffer_tree_ref(struct extent_buffer *eb)
4765 /* the ref bit is tricky. We have to make sure it is set
4766 * if we have the buffer dirty. Otherwise the
4767 * code to free a buffer can end up dropping a dirty
4770 * Once the ref bit is set, it won't go away while the
4771 * buffer is dirty or in writeback, and it also won't
4772 * go away while we have the reference count on the
4775 * We can't just set the ref bit without bumping the
4776 * ref on the eb because free_extent_buffer might
4777 * see the ref bit and try to clear it. If this happens
4778 * free_extent_buffer might end up dropping our original
4779 * ref by mistake and freeing the page before we are able
4780 * to add one more ref.
4782 * So bump the ref count first, then set the bit. If someone
4783 * beat us to it, drop the ref we added.
4785 refs = atomic_read(&eb->refs);
4786 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4789 spin_lock(&eb->refs_lock);
4790 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4791 atomic_inc(&eb->refs);
4792 spin_unlock(&eb->refs_lock);
4795 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4796 struct page *accessed)
4800 check_buffer_tree_ref(eb);
4802 num_pages = num_extent_pages(eb);
4803 for (i = 0; i < num_pages; i++) {
4804 struct page *p = eb->pages[i];
4807 mark_page_accessed(p);
4811 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4814 struct extent_buffer *eb;
4817 eb = radix_tree_lookup(&fs_info->buffer_radix,
4818 start >> PAGE_SHIFT);
4819 if (eb && atomic_inc_not_zero(&eb->refs)) {
4822 * Lock our eb's refs_lock to avoid races with
4823 * free_extent_buffer. When we get our eb it might be flagged
4824 * with EXTENT_BUFFER_STALE and another task running
4825 * free_extent_buffer might have seen that flag set,
4826 * eb->refs == 2, that the buffer isn't under IO (dirty and
4827 * writeback flags not set) and it's still in the tree (flag
4828 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4829 * of decrementing the extent buffer's reference count twice.
4830 * So here we could race and increment the eb's reference count,
4831 * clear its stale flag, mark it as dirty and drop our reference
4832 * before the other task finishes executing free_extent_buffer,
4833 * which would later result in an attempt to free an extent
4834 * buffer that is dirty.
4836 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4837 spin_lock(&eb->refs_lock);
4838 spin_unlock(&eb->refs_lock);
4840 mark_extent_buffer_accessed(eb, NULL);
4848 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4849 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4852 struct extent_buffer *eb, *exists = NULL;
4855 eb = find_extent_buffer(fs_info, start);
4858 eb = alloc_dummy_extent_buffer(fs_info, start);
4861 eb->fs_info = fs_info;
4863 ret = radix_tree_preload(GFP_NOFS);
4866 spin_lock(&fs_info->buffer_lock);
4867 ret = radix_tree_insert(&fs_info->buffer_radix,
4868 start >> PAGE_SHIFT, eb);
4869 spin_unlock(&fs_info->buffer_lock);
4870 radix_tree_preload_end();
4871 if (ret == -EEXIST) {
4872 exists = find_extent_buffer(fs_info, start);
4878 check_buffer_tree_ref(eb);
4879 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4883 btrfs_release_extent_buffer(eb);
4888 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4891 unsigned long len = fs_info->nodesize;
4894 unsigned long index = start >> PAGE_SHIFT;
4895 struct extent_buffer *eb;
4896 struct extent_buffer *exists = NULL;
4898 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4902 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4903 btrfs_err(fs_info, "bad tree block start %llu", start);
4904 return ERR_PTR(-EINVAL);
4907 eb = find_extent_buffer(fs_info, start);
4911 eb = __alloc_extent_buffer(fs_info, start, len);
4913 return ERR_PTR(-ENOMEM);
4915 num_pages = num_extent_pages(eb);
4916 for (i = 0; i < num_pages; i++, index++) {
4917 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4919 exists = ERR_PTR(-ENOMEM);
4923 spin_lock(&mapping->private_lock);
4924 if (PagePrivate(p)) {
4926 * We could have already allocated an eb for this page
4927 * and attached one so lets see if we can get a ref on
4928 * the existing eb, and if we can we know it's good and
4929 * we can just return that one, else we know we can just
4930 * overwrite page->private.
4932 exists = (struct extent_buffer *)p->private;
4933 if (atomic_inc_not_zero(&exists->refs)) {
4934 spin_unlock(&mapping->private_lock);
4937 mark_extent_buffer_accessed(exists, p);
4943 * Do this so attach doesn't complain and we need to
4944 * drop the ref the old guy had.
4946 ClearPagePrivate(p);
4947 WARN_ON(PageDirty(p));
4950 attach_extent_buffer_page(eb, p);
4951 spin_unlock(&mapping->private_lock);
4952 WARN_ON(PageDirty(p));
4954 if (!PageUptodate(p))
4958 * We can't unlock the pages just yet since the extent buffer
4959 * hasn't been properly inserted in the radix tree, this
4960 * opens a race with btree_releasepage which can free a page
4961 * while we are still filling in all pages for the buffer and
4966 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4968 ret = radix_tree_preload(GFP_NOFS);
4970 exists = ERR_PTR(ret);
4974 spin_lock(&fs_info->buffer_lock);
4975 ret = radix_tree_insert(&fs_info->buffer_radix,
4976 start >> PAGE_SHIFT, eb);
4977 spin_unlock(&fs_info->buffer_lock);
4978 radix_tree_preload_end();
4979 if (ret == -EEXIST) {
4980 exists = find_extent_buffer(fs_info, start);
4986 /* add one reference for the tree */
4987 check_buffer_tree_ref(eb);
4988 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4991 * Now it's safe to unlock the pages because any calls to
4992 * btree_releasepage will correctly detect that a page belongs to a
4993 * live buffer and won't free them prematurely.
4995 for (i = 0; i < num_pages; i++)
4996 unlock_page(eb->pages[i]);
5000 WARN_ON(!atomic_dec_and_test(&eb->refs));
5001 for (i = 0; i < num_pages; i++) {
5003 unlock_page(eb->pages[i]);
5006 btrfs_release_extent_buffer(eb);
5010 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5012 struct extent_buffer *eb =
5013 container_of(head, struct extent_buffer, rcu_head);
5015 __free_extent_buffer(eb);
5018 static int release_extent_buffer(struct extent_buffer *eb)
5020 lockdep_assert_held(&eb->refs_lock);
5022 WARN_ON(atomic_read(&eb->refs) == 0);
5023 if (atomic_dec_and_test(&eb->refs)) {
5024 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5025 struct btrfs_fs_info *fs_info = eb->fs_info;
5027 spin_unlock(&eb->refs_lock);
5029 spin_lock(&fs_info->buffer_lock);
5030 radix_tree_delete(&fs_info->buffer_radix,
5031 eb->start >> PAGE_SHIFT);
5032 spin_unlock(&fs_info->buffer_lock);
5034 spin_unlock(&eb->refs_lock);
5037 /* Should be safe to release our pages at this point */
5038 btrfs_release_extent_buffer_pages(eb);
5039 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5040 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
5041 __free_extent_buffer(eb);
5045 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5048 spin_unlock(&eb->refs_lock);
5053 void free_extent_buffer(struct extent_buffer *eb)
5061 refs = atomic_read(&eb->refs);
5062 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
5063 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
5066 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5071 spin_lock(&eb->refs_lock);
5072 if (atomic_read(&eb->refs) == 2 &&
5073 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5074 !extent_buffer_under_io(eb) &&
5075 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5076 atomic_dec(&eb->refs);
5079 * I know this is terrible, but it's temporary until we stop tracking
5080 * the uptodate bits and such for the extent buffers.
5082 release_extent_buffer(eb);
5085 void free_extent_buffer_stale(struct extent_buffer *eb)
5090 spin_lock(&eb->refs_lock);
5091 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5093 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5094 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5095 atomic_dec(&eb->refs);
5096 release_extent_buffer(eb);
5099 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5105 num_pages = num_extent_pages(eb);
5107 for (i = 0; i < num_pages; i++) {
5108 page = eb->pages[i];
5109 if (!PageDirty(page))
5113 WARN_ON(!PagePrivate(page));
5115 clear_page_dirty_for_io(page);
5116 xa_lock_irq(&page->mapping->i_pages);
5117 if (!PageDirty(page))
5118 __xa_clear_mark(&page->mapping->i_pages,
5119 page_index(page), PAGECACHE_TAG_DIRTY);
5120 xa_unlock_irq(&page->mapping->i_pages);
5121 ClearPageError(page);
5124 WARN_ON(atomic_read(&eb->refs) == 0);
5127 bool set_extent_buffer_dirty(struct extent_buffer *eb)
5133 check_buffer_tree_ref(eb);
5135 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5137 num_pages = num_extent_pages(eb);
5138 WARN_ON(atomic_read(&eb->refs) == 0);
5139 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5142 for (i = 0; i < num_pages; i++)
5143 set_page_dirty(eb->pages[i]);
5145 #ifdef CONFIG_BTRFS_DEBUG
5146 for (i = 0; i < num_pages; i++)
5147 ASSERT(PageDirty(eb->pages[i]));
5153 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5159 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5160 num_pages = num_extent_pages(eb);
5161 for (i = 0; i < num_pages; i++) {
5162 page = eb->pages[i];
5164 ClearPageUptodate(page);
5168 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5174 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5175 num_pages = num_extent_pages(eb);
5176 for (i = 0; i < num_pages; i++) {
5177 page = eb->pages[i];
5178 SetPageUptodate(page);
5182 int read_extent_buffer_pages(struct extent_io_tree *tree,
5183 struct extent_buffer *eb, int wait, int mirror_num)
5189 int locked_pages = 0;
5190 int all_uptodate = 1;
5192 unsigned long num_reads = 0;
5193 struct bio *bio = NULL;
5194 unsigned long bio_flags = 0;
5196 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5199 num_pages = num_extent_pages(eb);
5200 for (i = 0; i < num_pages; i++) {
5201 page = eb->pages[i];
5202 if (wait == WAIT_NONE) {
5203 if (!trylock_page(page))
5211 * We need to firstly lock all pages to make sure that
5212 * the uptodate bit of our pages won't be affected by
5213 * clear_extent_buffer_uptodate().
5215 for (i = 0; i < num_pages; i++) {
5216 page = eb->pages[i];
5217 if (!PageUptodate(page)) {
5224 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5228 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5229 eb->read_mirror = 0;
5230 atomic_set(&eb->io_pages, num_reads);
5231 for (i = 0; i < num_pages; i++) {
5232 page = eb->pages[i];
5234 if (!PageUptodate(page)) {
5236 atomic_dec(&eb->io_pages);
5241 ClearPageError(page);
5242 err = __extent_read_full_page(tree, page,
5243 btree_get_extent, &bio,
5244 mirror_num, &bio_flags,
5249 * We use &bio in above __extent_read_full_page,
5250 * so we ensure that if it returns error, the
5251 * current page fails to add itself to bio and
5252 * it's been unlocked.
5254 * We must dec io_pages by ourselves.
5256 atomic_dec(&eb->io_pages);
5264 err = submit_one_bio(bio, mirror_num, bio_flags);
5269 if (ret || wait != WAIT_COMPLETE)
5272 for (i = 0; i < num_pages; i++) {
5273 page = eb->pages[i];
5274 wait_on_page_locked(page);
5275 if (!PageUptodate(page))
5282 while (locked_pages > 0) {
5284 page = eb->pages[locked_pages];
5290 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5291 unsigned long start, unsigned long len)
5297 char *dst = (char *)dstv;
5298 size_t start_offset = offset_in_page(eb->start);
5299 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5301 if (start + len > eb->len) {
5302 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5303 eb->start, eb->len, start, len);
5304 memset(dst, 0, len);
5308 offset = offset_in_page(start_offset + start);
5311 page = eb->pages[i];
5313 cur = min(len, (PAGE_SIZE - offset));
5314 kaddr = page_address(page);
5315 memcpy(dst, kaddr + offset, cur);
5324 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5326 unsigned long start, unsigned long len)
5332 char __user *dst = (char __user *)dstv;
5333 size_t start_offset = offset_in_page(eb->start);
5334 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5337 WARN_ON(start > eb->len);
5338 WARN_ON(start + len > eb->start + eb->len);
5340 offset = offset_in_page(start_offset + start);
5343 page = eb->pages[i];
5345 cur = min(len, (PAGE_SIZE - offset));
5346 kaddr = page_address(page);
5347 if (copy_to_user(dst, kaddr + offset, cur)) {
5362 * return 0 if the item is found within a page.
5363 * return 1 if the item spans two pages.
5364 * return -EINVAL otherwise.
5366 int map_private_extent_buffer(const struct extent_buffer *eb,
5367 unsigned long start, unsigned long min_len,
5368 char **map, unsigned long *map_start,
5369 unsigned long *map_len)
5374 size_t start_offset = offset_in_page(eb->start);
5375 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5376 unsigned long end_i = (start_offset + start + min_len - 1) >>
5379 if (start + min_len > eb->len) {
5380 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5381 eb->start, eb->len, start, min_len);
5389 offset = start_offset;
5393 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5397 kaddr = page_address(p);
5398 *map = kaddr + offset;
5399 *map_len = PAGE_SIZE - offset;
5403 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5404 unsigned long start, unsigned long len)
5410 char *ptr = (char *)ptrv;
5411 size_t start_offset = offset_in_page(eb->start);
5412 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5415 WARN_ON(start > eb->len);
5416 WARN_ON(start + len > eb->start + eb->len);
5418 offset = offset_in_page(start_offset + start);
5421 page = eb->pages[i];
5423 cur = min(len, (PAGE_SIZE - offset));
5425 kaddr = page_address(page);
5426 ret = memcmp(ptr, kaddr + offset, cur);
5438 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5443 WARN_ON(!PageUptodate(eb->pages[0]));
5444 kaddr = page_address(eb->pages[0]);
5445 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5449 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5453 WARN_ON(!PageUptodate(eb->pages[0]));
5454 kaddr = page_address(eb->pages[0]);
5455 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5459 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5460 unsigned long start, unsigned long len)
5466 char *src = (char *)srcv;
5467 size_t start_offset = offset_in_page(eb->start);
5468 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5470 WARN_ON(start > eb->len);
5471 WARN_ON(start + len > eb->start + eb->len);
5473 offset = offset_in_page(start_offset + start);
5476 page = eb->pages[i];
5477 WARN_ON(!PageUptodate(page));
5479 cur = min(len, PAGE_SIZE - offset);
5480 kaddr = page_address(page);
5481 memcpy(kaddr + offset, src, cur);
5490 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5497 size_t start_offset = offset_in_page(eb->start);
5498 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5500 WARN_ON(start > eb->len);
5501 WARN_ON(start + len > eb->start + eb->len);
5503 offset = offset_in_page(start_offset + start);
5506 page = eb->pages[i];
5507 WARN_ON(!PageUptodate(page));
5509 cur = min(len, PAGE_SIZE - offset);
5510 kaddr = page_address(page);
5511 memset(kaddr + offset, 0, cur);
5519 void copy_extent_buffer_full(struct extent_buffer *dst,
5520 struct extent_buffer *src)
5525 ASSERT(dst->len == src->len);
5527 num_pages = num_extent_pages(dst);
5528 for (i = 0; i < num_pages; i++)
5529 copy_page(page_address(dst->pages[i]),
5530 page_address(src->pages[i]));
5533 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5534 unsigned long dst_offset, unsigned long src_offset,
5537 u64 dst_len = dst->len;
5542 size_t start_offset = offset_in_page(dst->start);
5543 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5545 WARN_ON(src->len != dst_len);
5547 offset = offset_in_page(start_offset + dst_offset);
5550 page = dst->pages[i];
5551 WARN_ON(!PageUptodate(page));
5553 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5555 kaddr = page_address(page);
5556 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5566 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5568 * @eb: the extent buffer
5569 * @start: offset of the bitmap item in the extent buffer
5571 * @page_index: return index of the page in the extent buffer that contains the
5573 * @page_offset: return offset into the page given by page_index
5575 * This helper hides the ugliness of finding the byte in an extent buffer which
5576 * contains a given bit.
5578 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5579 unsigned long start, unsigned long nr,
5580 unsigned long *page_index,
5581 size_t *page_offset)
5583 size_t start_offset = offset_in_page(eb->start);
5584 size_t byte_offset = BIT_BYTE(nr);
5588 * The byte we want is the offset of the extent buffer + the offset of
5589 * the bitmap item in the extent buffer + the offset of the byte in the
5592 offset = start_offset + start + byte_offset;
5594 *page_index = offset >> PAGE_SHIFT;
5595 *page_offset = offset_in_page(offset);
5599 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5600 * @eb: the extent buffer
5601 * @start: offset of the bitmap item in the extent buffer
5602 * @nr: bit number to test
5604 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5612 eb_bitmap_offset(eb, start, nr, &i, &offset);
5613 page = eb->pages[i];
5614 WARN_ON(!PageUptodate(page));
5615 kaddr = page_address(page);
5616 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5620 * extent_buffer_bitmap_set - set an area of a bitmap
5621 * @eb: the extent buffer
5622 * @start: offset of the bitmap item in the extent buffer
5623 * @pos: bit number of the first bit
5624 * @len: number of bits to set
5626 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5627 unsigned long pos, unsigned long len)
5633 const unsigned int size = pos + len;
5634 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5635 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5637 eb_bitmap_offset(eb, start, pos, &i, &offset);
5638 page = eb->pages[i];
5639 WARN_ON(!PageUptodate(page));
5640 kaddr = page_address(page);
5642 while (len >= bits_to_set) {
5643 kaddr[offset] |= mask_to_set;
5645 bits_to_set = BITS_PER_BYTE;
5647 if (++offset >= PAGE_SIZE && len > 0) {
5649 page = eb->pages[++i];
5650 WARN_ON(!PageUptodate(page));
5651 kaddr = page_address(page);
5655 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5656 kaddr[offset] |= mask_to_set;
5662 * extent_buffer_bitmap_clear - clear an area of a bitmap
5663 * @eb: the extent buffer
5664 * @start: offset of the bitmap item in the extent buffer
5665 * @pos: bit number of the first bit
5666 * @len: number of bits to clear
5668 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5669 unsigned long pos, unsigned long len)
5675 const unsigned int size = pos + len;
5676 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5677 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5679 eb_bitmap_offset(eb, start, pos, &i, &offset);
5680 page = eb->pages[i];
5681 WARN_ON(!PageUptodate(page));
5682 kaddr = page_address(page);
5684 while (len >= bits_to_clear) {
5685 kaddr[offset] &= ~mask_to_clear;
5686 len -= bits_to_clear;
5687 bits_to_clear = BITS_PER_BYTE;
5689 if (++offset >= PAGE_SIZE && len > 0) {
5691 page = eb->pages[++i];
5692 WARN_ON(!PageUptodate(page));
5693 kaddr = page_address(page);
5697 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5698 kaddr[offset] &= ~mask_to_clear;
5702 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5704 unsigned long distance = (src > dst) ? src - dst : dst - src;
5705 return distance < len;
5708 static void copy_pages(struct page *dst_page, struct page *src_page,
5709 unsigned long dst_off, unsigned long src_off,
5712 char *dst_kaddr = page_address(dst_page);
5714 int must_memmove = 0;
5716 if (dst_page != src_page) {
5717 src_kaddr = page_address(src_page);
5719 src_kaddr = dst_kaddr;
5720 if (areas_overlap(src_off, dst_off, len))
5725 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5727 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5730 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5731 unsigned long src_offset, unsigned long len)
5733 struct btrfs_fs_info *fs_info = dst->fs_info;
5735 size_t dst_off_in_page;
5736 size_t src_off_in_page;
5737 size_t start_offset = offset_in_page(dst->start);
5738 unsigned long dst_i;
5739 unsigned long src_i;
5741 if (src_offset + len > dst->len) {
5743 "memmove bogus src_offset %lu move len %lu dst len %lu",
5744 src_offset, len, dst->len);
5747 if (dst_offset + len > dst->len) {
5749 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5750 dst_offset, len, dst->len);
5755 dst_off_in_page = offset_in_page(start_offset + dst_offset);
5756 src_off_in_page = offset_in_page(start_offset + src_offset);
5758 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5759 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5761 cur = min(len, (unsigned long)(PAGE_SIZE -
5763 cur = min_t(unsigned long, cur,
5764 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5766 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5767 dst_off_in_page, src_off_in_page, cur);
5775 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5776 unsigned long src_offset, unsigned long len)
5778 struct btrfs_fs_info *fs_info = dst->fs_info;
5780 size_t dst_off_in_page;
5781 size_t src_off_in_page;
5782 unsigned long dst_end = dst_offset + len - 1;
5783 unsigned long src_end = src_offset + len - 1;
5784 size_t start_offset = offset_in_page(dst->start);
5785 unsigned long dst_i;
5786 unsigned long src_i;
5788 if (src_offset + len > dst->len) {
5790 "memmove bogus src_offset %lu move len %lu len %lu",
5791 src_offset, len, dst->len);
5794 if (dst_offset + len > dst->len) {
5796 "memmove bogus dst_offset %lu move len %lu len %lu",
5797 dst_offset, len, dst->len);
5800 if (dst_offset < src_offset) {
5801 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5805 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5806 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5808 dst_off_in_page = offset_in_page(start_offset + dst_end);
5809 src_off_in_page = offset_in_page(start_offset + src_end);
5811 cur = min_t(unsigned long, len, src_off_in_page + 1);
5812 cur = min(cur, dst_off_in_page + 1);
5813 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5814 dst_off_in_page - cur + 1,
5815 src_off_in_page - cur + 1, cur);
5823 int try_release_extent_buffer(struct page *page)
5825 struct extent_buffer *eb;
5828 * We need to make sure nobody is attaching this page to an eb right
5831 spin_lock(&page->mapping->private_lock);
5832 if (!PagePrivate(page)) {
5833 spin_unlock(&page->mapping->private_lock);
5837 eb = (struct extent_buffer *)page->private;
5841 * This is a little awful but should be ok, we need to make sure that
5842 * the eb doesn't disappear out from under us while we're looking at
5845 spin_lock(&eb->refs_lock);
5846 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5847 spin_unlock(&eb->refs_lock);
5848 spin_unlock(&page->mapping->private_lock);
5851 spin_unlock(&page->mapping->private_lock);
5854 * If tree ref isn't set then we know the ref on this eb is a real ref,
5855 * so just return, this page will likely be freed soon anyway.
5857 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5858 spin_unlock(&eb->refs_lock);
5862 return release_extent_buffer(eb);